versions/1.5.0/api/python/gluon/gluon.html

<!DOCTYPE html> <html lang="en"> <head> <meta charset="utf-8"/> <meta content="IE=edge" http-equiv="X-UA-Compatible"/> <meta content="width=device-width, initial-scale=1" name="viewport"/> <meta content="Gluon Package" property="og:title"> <meta content="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/og-logo.png" property="og:image"> <meta content="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/og-logo.png" property="og:image:secure_url"> <meta content="Gluon Package" property="og:description"/> <title>Gluon Package — mxnet documentation</title> <link crossorigin="anonymous" href="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.6/css/bootstrap.min.css" integrity="sha384-1q8mTJOASx8j1Au+a5WDVnPi2lkFfwwEAa8hDDdjZlpLegxhjVME1fgjWPGmkzs7" rel="stylesheet"/> <link href="https://maxcdn.bootstrapcdn.com/font-awesome/4.5.0/css/font-awesome.min.css" rel="stylesheet"/> <link href="../../../_static/basic.css" rel="stylesheet" type="text/css"> <link href="../../../_static/pygments.css" rel="stylesheet" type="text/css"> <link href="../../../_static/mxnet.css" rel="stylesheet" type="text/css"/> <script type="text/javascript"> var DOCUMENTATION_OPTIONS = { URL_ROOT: '../../../', VERSION: '', COLLAPSE_INDEX: false, FILE_SUFFIX: '.html', HAS_SOURCE: true, SOURCELINK_SUFFIX: '.txt' }; </script> <script src="https://code.jquery.com/jquery-1.11.1.min.js" type="text/javascript"></script> <script src="../../../_static/underscore.js" type="text/javascript"></script> <script src="../../../_static/searchtools_custom.js" type="text/javascript"></script> <script src="../../../_static/doctools.js" type="text/javascript"></script> <script src="../../../_static/selectlang.js" type="text/javascript"></script> <script src="https://cdnjs.cloudflare.com/ajax/libs/mathjax/2.7.1/MathJax.js?config=TeX-AMS-MML_HTMLorMML" type="text/javascript"></script> <script type="text/javascript"> jQuery(function() { Search.loadIndex("/versions/1.5.0/searchindex.js"); Search.init();}); </script>          <link href="../../../genindex.html" rel="index" title="Index"> <link href="../../../search.html" rel="search" title="Search"/> <link href="../index.html" rel="up" title="MXNet - Python API"/> <link href="nn.html" rel="next" title="Gluon Neural Network Layers"/> <link href="../contrib/svrg_optimization.html" rel="prev" title="SVRG Optimization in Python Module API"/> <link href="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/mxnet-icon.png" rel="icon" type="image/png"/> </link></link></link></meta></meta></meta></head> <body background="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/mxnet-background-compressed.jpeg" role="document"> <div class="content-block"><div class="navbar navbar-fixed-top"> <div class="container" id="navContainer"> <div class="innder" id="header-inner"> <h1 id="logo-wrap"> <a href="../../../" id="logo"><img src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/mxnet_logo.png"/></a> </h1> <nav class="nav-bar" id="main-nav"> <a class="main-nav-link" href="/versions/1.5.0/install/index.html">Install</a> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">Gluon </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu"> <li><a class="main-nav-link" href="/versions/1.5.0/tutorials/gluon/gluon.html">About</a></li> <li><a class="main-nav-link" href="https://www.d2l.ai/">Dive into Deep Learning</a></li> <li><a class="main-nav-link" href="https://gluon-cv.mxnet.io">GluonCV Toolkit</a></li> <li><a class="main-nav-link" href="https://gluon-nlp.mxnet.io/">GluonNLP Toolkit</a></li> </ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">API </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu"> <li><a class="main-nav-link" href="/versions/1.5.0/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/clojure/index.html">Clojure</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/java/index.html">Java</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/scala/index.html">Scala</a></li> </ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">Docs </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu-docs"> <li><a class="main-nav-link" href="/versions/1.5.0/faq/index.html">FAQ</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/tutorials/index.html">Tutorials</a> <li><a class="main-nav-link" href="https://github.com/apache/incubator-mxnet/tree/1.5.0/example">Examples</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/architecture/index.html">Architecture</a></li> <li><a class="main-nav-link" href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home">Developer Wiki</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/model_zoo/index.html">Model Zoo</a></li> <li><a class="main-nav-link" href="https://github.com/onnx/onnx-mxnet">ONNX</a></li> </li></ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">Community </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu-community"> <li><a class="main-nav-link" href="http://discuss.mxnet.io">Forum</a></li> <li><a class="main-nav-link" href="https://github.com/apache/incubator-mxnet/tree/1.5.0">Github</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/community/contribute.html">Contribute</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/community/ecosystem.html">Ecosystem</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/community/powered_by.html">Powered By</a></li> </ul> <a href="#" class="main-nav-link dropdown-toggle" data-toggle="dropdown" role="button" aria-haspopup="true" aria-expanded="true">1.5.0</a><ul id="package-dropdown-menu" class="dropdown-menu"><li><a href="/">master</a></li><li><a href="/versions/1.7.0/">1.7.0</a></li><li><a href=/versions/1.6.0/>1.6.0</a></li><li><a href=/versions/1.5.0/>1.5.0</a></li><li><a href=/versions/1.4.1/>1.4.1</a></li><li><a href=/versions/1.3.1/>1.3.1</a></li><li><a href=/versions/1.2.1/>1.2.1</a></li><li><a href=/versions/1.1.0/>1.1.0</a></li><li><a href=/versions/1.0.0/>1.0.0</a></li><li><a href=/versions/0.12.1/>0.12.1</a></li><li><a href=/versions/0.11.0/>0.11.0</a></li></ul></nav> <script> function getRootPath(){ return "../../../" } </script> <div class="burgerIcon dropdown"> <a class="dropdown-toggle" data-toggle="dropdown" href="#" role="button">☰</a> <ul class="dropdown-menu" id="burgerMenu"> <li><a href="/versions/1.5.0/install/index.html">Install</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/tutorials/index.html">Tutorials</a></li> <li class="dropdown-submenu dropdown"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">Gluon</a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu"> <li><a class="main-nav-link" href="/versions/1.5.0/tutorials/gluon/gluon.html">About</a></li> <li><a class="main-nav-link" href="http://gluon.mxnet.io">The Straight Dope (Tutorials)</a></li> <li><a class="main-nav-link" href="https://gluon-cv.mxnet.io">GluonCV Toolkit</a></li> <li><a class="main-nav-link" href="https://gluon-nlp.mxnet.io/">GluonNLP Toolkit</a></li> </ul> </li> <li class="dropdown-submenu"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">API</a> <ul class="dropdown-menu"> <li><a class="main-nav-link" href="/versions/1.5.0/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/clojure/index.html">Clojure</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/java/index.html">Java</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/1.5.0/api/scala/index.html">Scala</a></li> </ul> </li> <li class="dropdown-submenu"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">Docs</a> <ul class="dropdown-menu"> <li><a href="/versions/1.5.0/faq/index.html" tabindex="-1">FAQ</a></li> <li><a href="/versions/1.5.0/tutorials/index.html" tabindex="-1">Tutorials</a></li> <li><a href="https://github.com/apache/incubator-mxnet/tree/1.5.0/example" tabindex="-1">Examples</a></li> <li><a href="/versions/1.5.0/architecture/index.html" tabindex="-1">Architecture</a></li> <li><a href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home" tabindex="-1">Developer Wiki</a></li> <li><a href="/versions/1.5.0/model_zoo/index.html" tabindex="-1">Gluon Model Zoo</a></li> <li><a href="https://github.com/onnx/onnx-mxnet" tabindex="-1">ONNX</a></li> </ul> </li> <li class="dropdown-submenu dropdown"> <a aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" role="button" tabindex="-1">Community</a> <ul class="dropdown-menu"> <li><a href="http://discuss.mxnet.io" tabindex="-1">Forum</a></li> <li><a href="https://github.com/apache/incubator-mxnet/tree/1.5.0" tabindex="-1">Github</a></li> <li><a href="/versions/1.5.0/community/contribute.html" tabindex="-1">Contribute</a></li> <li><a href="/versions/1.5.0/community/ecosystem.html" tabindex="-1">Ecosystem</a></li> <li><a href="/versions/1.5.0/community/powered_by.html" tabindex="-1">Powered By</a></li> </ul> </li> <li id="dropdown-menu-position-anchor-version-mobile" class="dropdown-submenu" style="position: relative"><a href="#" tabindex="-1">1.5.0</a><ul class="dropdown-menu"><li><a tabindex="-1" href=/>master</a></li><li><a tabindex="-1" href=/versions/1.6.0/>1.6.0</a></li><li><a tabindex="-1" href=/versions/1.5.0/>1.5.0</a></li><li><a tabindex="-1" href=/versions/1.4.1/>1.4.1</a></li><li><a tabindex="-1" href=/versions/1.3.1/>1.3.1</a></li><li><a tabindex="-1" href=/versions/1.2.1/>1.2.1</a></li><li><a tabindex="-1" href=/versions/1.1.0/>1.1.0</a></li><li><a tabindex="-1" href=/versions/1.0.0/>1.0.0</a></li><li><a tabindex="-1" href=/versions/0.12.1/>0.12.1</a></li><li><a tabindex="-1" href=/versions/0.11.0/>0.11.0</a></li></ul></li></ul> </div> <div class="plusIcon dropdown"> <a class="dropdown-toggle" data-toggle="dropdown" href="#" role="button"></a> <ul class="dropdown-menu dropdown-menu-right" id="plusMenu"></ul> </div> <div id="search-input-wrap"> <form action="../../../search.html" autocomplete="off" class="" method="get" role="search"> <div class="form-group inner-addon left-addon"> <input class="form-control" name="q" placeholder="Search" type="text"/> </div> <input name="check_keywords" type="hidden" value="yes"> <input name="area" type="hidden" value="default"/> </input></form> <div id="search-preview"></div> </div> <div id="searchIcon"> </div>            </div> </div> </div> <script type="text/javascript"> $('body').css('background', 'white'); </script> <div class="container"> <div class="row"> <div aria-label="main navigation" class="sphinxsidebar leftsidebar" role="navigation"> <div class="sphinxsidebarwrapper"> <ul> <li class="toctree-l1"><a class="reference internal" href="../../index.html">MXNet APIs</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../architecture/index.html">MXNet Architecture</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../community/index.html">MXNet Community</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../faq/index.html">MXNet FAQ</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../gluon/index.html">About Gluon</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../install/index.html">Installing MXNet</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../install/index.html#nvidia-jetson-tx-family">Nvidia Jetson TX family</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../install/index.html#source-download">Source Download</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../model_zoo/index.html">MXNet Model Zoo</a></li> <li class="toctree-l1"><a class="reference internal" href="../../../tutorials/index.html">Tutorials</a></li> </ul> </div> </div> <div class="content"> <div class="page-tracker"></div>            <div class="section" id="gluon-package"> <h1>Gluon Package<a class="headerlink" href="#gluon-package" title="Permalink to this headline">¶</a></h1> <script src="../../../_static/js/auto_module_index.js" type="text/javascript"></script><div class="section" id="overview"> <h2>Overview<a class="headerlink" href="#overview" title="Permalink to this headline">¶</a></h2> The Gluon package is a high-level interface for MXNet designed to be easy to use, while keeping most of the flexibility of a low level API. Gluon supports both imperative and symbolic programming, making it easy to train complex models imperatively in Python and then deploy with a symbolic graph in C++ and Scala. Based on the the <a class="reference external" href="https://github.com/gluon-api/gluon-api">Gluon API specification</a>, the Gluon API in Apache MXNet provides a clear, concise, and simple API for deep learning. It makes it easy to prototype, build, and train deep learning models without sacrificing training speed. Advantages <ol class="simple"> <li>Simple, Easy-to-Understand Code: Gluon offers a full set of plug-and-play neural network building blocks, including predefined layers, optimizers, and initializers.</li> <li>Flexible, Imperative Structure: Gluon does not require the neural network model to be rigidly defined, but rather brings the training algorithm and model closer together to provide flexibility in the development process.</li> <li>Dynamic Graphs: Gluon enables developers to define neural network models that are dynamic, meaning they can be built on the fly, with any structure, and using any of Python’s native control flow.</li> <li>High Performance: Gluon provides all of the above benefits without impacting the training speed that the underlying engine provides.</li> </ol> Examples Simple, Easy-to-Understand Code Use plug-and-play neural network building blocks, including predefined layers, optimizers, and initializers: <div class="highlight-default"><div class="highlight"><pre>net = gluon.nn.Sequential() # When instantiated, Sequential stores a chain of neural network layers. # Once presented with data, Sequential executes each layer in turn, using # the output of one layer as the input for the next with net.name_scope(): net.add(gluon.nn.Dense(256, activation="relu")) # 1st layer (256 nodes) net.add(gluon.nn.Dense(256, activation="relu")) # 2nd hidden layer net.add(gluon.nn.Dense(num_outputs)) </pre></div> </div> Flexible, Imperative Structure Prototype, build, and train neural networks in fully imperative manner using the MXNet autograd package and the Gluon trainer method: <div class="highlight-default"><div class="highlight"><pre>epochs = 10 for e in range(epochs): for i, (data, label) in enumerate(train_data): with autograd.record(): output = net(data) # the forward iteration loss = softmax_cross_entropy(output, label) loss.backward() trainer.step(data.shape[0]) </pre></div> </div> Dynamic Graphs Build neural networks on the fly for use cases where neural networks must change in size and shape during model training: <div class="highlight-default"><div class="highlight"><pre>def forward(self, F, inputs, tree): children_outputs = [self.forward(F, inputs, child) for child in tree.children] #Recursively builds the neural network based on each input sentence’s #syntactic structure during the model definition and training process ... </pre></div> </div> High Performance Easily cache the neural network to achieve high performance by defining your neural network with HybridSequential and calling the hybridize method: <div class="highlight-default"><div class="highlight"><pre>net = nn.HybridSequential() with net.name_scope(): net.add(nn.Dense(256, activation="relu")) net.add(nn.Dense(128, activation="relu")) net.add(nn.Dense(2)) net.hybridize() </pre></div> </div> </div> <div class="section" id="contents"> <h2>Contents<a class="headerlink" href="#contents" title="Permalink to this headline">¶</a></h2> <div class="toctree-wrapper compound"> <ul> <li class="toctree-l1"><a class="reference internal" href="nn.html">Gluon Neural Network Layers</a></li> <li class="toctree-l1"><a class="reference internal" href="rnn.html">Gluon Recurrent Neural Network API</a></li> <li class="toctree-l1"><a class="reference internal" href="loss.html">Gluon Loss API</a></li> <li class="toctree-l1"><a class="reference internal" href="data.html">Gluon Data API</a></li> <li class="toctree-l1"><a class="reference internal" href="model_zoo.html">Gluon Model Zoo</a></li> <li class="toctree-l1"><a class="reference internal" href="contrib.html">Gluon Contrib API</a></li> </ul> </div> </div> <div class="section" id="parameter"> <h2>Parameter<a class="headerlink" href="#parameter" title="Permalink to this headline">¶</a></h2> <table border="1" class="longtable docutils"> <colgroup> <col width="10%"/> <col width="90%"/> </colgroup> <tbody valign="top"> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-obj docutils literal">Parameter</code></a></td> <td>A Container holding parameters (weights) of Blocks.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.Constant" title="mxnet.gluon.Constant"><code class="xref py py-obj docutils literal">Constant</code></a></td> <td>A constant parameter for holding immutable tensors.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict"><code class="xref py py-obj docutils literal">ParameterDict</code></a></td> <td>A dictionary managing a set of parameters.</td> </tr> </tbody> </table> </div> <div class="section" id="containers"> <h2>Containers<a class="headerlink" href="#containers" title="Permalink to this headline">¶</a></h2> <table border="1" class="longtable docutils"> <colgroup> <col width="10%"/> <col width="90%"/> </colgroup> <tbody valign="top"> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-obj docutils literal">Block</code></a></td> <td>Base class for all neural network layers and models.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.HybridBlock" title="mxnet.gluon.HybridBlock"><code class="xref py py-obj docutils literal">HybridBlock</code></a></td> <td><cite>HybridBlock</cite> supports forwarding with both Symbol and NDArray.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.SymbolBlock" title="mxnet.gluon.SymbolBlock"><code class="xref py py-obj docutils literal">SymbolBlock</code></a></td> <td>Construct block from symbol.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Sequential" title="mxnet.gluon.nn.Sequential"><code class="xref py py-obj docutils literal">nn.Sequential</code></a></td> <td>Stacks Blocks sequentially.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.HybridSequential" title="mxnet.gluon.nn.HybridSequential"><code class="xref py py-obj docutils literal">nn.HybridSequential</code></a></td> <td>Stacks HybridBlocks sequentially.</td> </tr> </tbody> </table> </div> <div class="section" id="trainer"> <h2>Trainer<a class="headerlink" href="#trainer" title="Permalink to this headline">¶</a></h2> <table border="1" class="longtable docutils"> <colgroup> <col width="10%"/> <col width="90%"/> </colgroup> <tbody valign="top"> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.Trainer" title="mxnet.gluon.Trainer"><code class="xref py py-obj docutils literal">Trainer</code></a></td> <td>Applies an <cite>Optimizer</cite> on a set of Parameters.</td> </tr> </tbody> </table> </div> <div class="section" id="utilities"> <h2>Utilities<a class="headerlink" href="#utilities" title="Permalink to this headline">¶</a></h2> <table border="1" class="longtable docutils"> <colgroup> <col width="10%"/> <col width="90%"/> </colgroup> <tbody valign="top"> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.utils.split_data" title="mxnet.gluon.utils.split_data"><code class="xref py py-obj docutils literal">split_data</code></a></td> <td>Splits an NDArray into <cite>num_slice</cite> slices along <cite>batch_axis</cite>.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.utils.split_and_load" title="mxnet.gluon.utils.split_and_load"><code class="xref py py-obj docutils literal">split_and_load</code></a></td> <td>Splits an NDArray into <cite>len(ctx_list)</cite> slices along <cite>batch_axis</cite> and loads each slice to one context in <cite>ctx_list</cite>.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.utils.clip_global_norm" title="mxnet.gluon.utils.clip_global_norm"><code class="xref py py-obj docutils literal">clip_global_norm</code></a></td> <td>Rescales NDArrays so that the sum of their 2-norm is smaller than <cite>max_norm</cite>.</td> </tr> </tbody> </table> </div> <div class="section" id="api-reference"> <h2>API Reference<a class="headerlink" href="#api-reference" title="Permalink to this headline">¶</a></h2> <script src="../../../_static/js/auto_module_index.js" type="text/javascript"></script>Neural network module. <dl class="class"> <dt id="mxnet.gluon.Block"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">Block</code>(prefix=None, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block">[source]</a><a class="headerlink" href="#mxnet.gluon.Block" title="Permalink to this definition">¶</a></dt> <dd>Base class for all neural network layers and models. Your models should subclass this class. <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> can be nested recursively in a tree structure. You can create and assign child <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> as regular attributes: <div class="highlight-default"><div class="highlight"><pre>from mxnet.gluon import Block, nn from mxnet import ndarray as F class Model(Block): def __init__(self, **kwargs): super(Model, self).__init__(**kwargs) # use name_scope to give child Blocks appropriate names. with self.name_scope(): self.dense0 = nn.Dense(20) self.dense1 = nn.Dense(20) def forward(self, x): x = F.relu(self.dense0(x)) return F.relu(self.dense1(x)) model = Model() model.initialize(ctx=mx.cpu(0)) model(F.zeros((10, 10), ctx=mx.cpu(0))) </pre></div> </div> Child <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> assigned this way will be registered and <a class="reference internal" href="#mxnet.gluon.Block.collect_params" title="mxnet.gluon.Block.collect_params"><code class="xref py py-meth docutils literal">collect_params()</code></a> will collect their Parameters recursively. You can also manually register child blocks with <a class="reference internal" href="#mxnet.gluon.Block.register_child" title="mxnet.gluon.Block.register_child"><code class="xref py py-meth docutils literal">register_child()</code></a>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>prefix (str) – Prefix acts like a name space. All children blocks created in parent block’s <a class="reference internal" href="#mxnet.gluon.Block.name_scope" title="mxnet.gluon.Block.name_scope"><code class="xref py py-meth docutils literal">name_scope()</code></a> will have parent block’s prefix in their name. Please refer to <a class="reference external" href="/versions/1.5.0/tutorials/gluon/naming.html">naming tutorial</a> for more info on prefix and naming.</li> <li>params (<a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict">ParameterDict</a> or None) – <a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict"><code class="xref py py-class docutils literal">ParameterDict</code></a> for sharing weights with the new <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a>. For example, if you want <code class="docutils literal">dense1</code> to share <code class="docutils literal">dense0</code>‘s weights, you can do: <div class="highlight-default"><div class="highlight"><pre>dense0 = nn.Dense(20) dense1 = nn.Dense(20, params=dense0.collect_params()) </pre></div> </div> </li> </ul> </td> </tr> </tbody> </table> <dl class="method"> <dt id="mxnet.gluon.Block.__call__"> <code class="descname">__call__</code>(*args)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.__call__">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.__call__" title="Permalink to this definition">¶</a></dt> <dd>Calls forward. Only accepts positional arguments. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.__setattr__"> <code class="descname">__setattr__</code>(name, value)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.__setattr__">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.__setattr__" title="Permalink to this definition">¶</a></dt> <dd>Registers parameters. </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Block.__weakref__"> <code class="descname">__weakref__</code><a class="headerlink" href="#mxnet.gluon.Block.__weakref__" title="Permalink to this definition">¶</a></dt> <dd>list of weak references to the object (if defined) </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.apply"> <code class="descname">apply</code>(fn)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.apply">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.apply" title="Permalink to this definition">¶</a></dt> <dd>Applies <code class="docutils literal">fn</code> recursively to every child block as well as self. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">fn (callable) – Function to be applied to each submodule, of form <cite>fn(block)</cite>.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">this block</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.cast"> <code class="descname">cast</code>(dtype)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.cast">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.cast" title="Permalink to this definition">¶</a></dt> <dd>Cast this Block to use another data type. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">dtype (str or numpy.dtype) – The new data type.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.collect_params"> <code class="descname">collect_params</code>(select=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.collect_params">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.collect_params" title="Permalink to this definition">¶</a></dt> <dd>Returns a <a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict"><code class="xref py py-class docutils literal">ParameterDict</code></a> containing this <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> and all of its children’s Parameters(default), also can returns the select <a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict"><code class="xref py py-class docutils literal">ParameterDict</code></a> which match some given regular expressions. For example, collect the specified parameters in [‘conv1_weight’, ‘conv1_bias’, ‘fc_weight’, ‘fc_bias’]: <div class="highlight-default"><div class="highlight"><pre>model.collect_params('conv1_weight|conv1_bias|fc_weight|fc_bias') </pre></div> </div> or collect all parameters whose names end with ‘weight’ or ‘bias’, this can be done using regular expressions: <div class="highlight-default"><div class="highlight"><pre>model.collect_params('.*weight|.*bias') </pre></div> </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">select (str) – regular expressions</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">The selected <a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict"><code class="xref py py-class docutils literal">ParameterDict</code></a></td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.forward"> <code class="descname">forward</code>(*args)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.forward">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.forward" title="Permalink to this definition">¶</a></dt> <dd>Overrides to implement forward computation using <code class="xref py py-class docutils literal">NDArray</code>. Only accepts positional arguments. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">*args (list of NDArray) – Input tensors.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.hybridize"> <code class="descname">hybridize</code>(active=True, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.hybridize">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.hybridize" title="Permalink to this definition">¶</a></dt> <dd>Activates or deactivates <a class="reference internal" href="#mxnet.gluon.HybridBlock" title="mxnet.gluon.HybridBlock"><code class="xref py py-class docutils literal">HybridBlock</code></a> s recursively. Has no effect on non-hybrid children. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>active (bool, default True) – Whether to turn hybrid on or off.</li> <li>static_alloc (bool, default False) – Statically allocate memory to improve speed. Memory usage may increase.</li> <li>static_shape (bool, default False) – Optimize for invariant input shapes between iterations. Must also set static_alloc to True. Change of input shapes is still allowed but slower.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.initialize"> <code class="descname">initialize</code>(init=<mxnet.initializer.Uniform object>, ctx=None, verbose=False, force_reinit=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.initialize">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.initialize" title="Permalink to this definition">¶</a></dt> <dd>Initializes <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a> s of this <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> and its children. Equivalent to <code class="docutils literal">block.collect_params().initialize(...)</code> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>init (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – Global default Initializer to be used when <code class="xref py py-meth docutils literal">Parameter.init()</code> is <code class="docutils literal">None</code>. Otherwise, <code class="xref py py-meth docutils literal">Parameter.init()</code> takes precedence.</li> <li>ctx (Context or list of Context) – Keeps a copy of Parameters on one or many context(s).</li> <li>verbose (bool, default False) – Whether to verbosely print out details on initialization.</li> <li>force_reinit (bool, default False) – Whether to force re-initialization if parameter is already initialized.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.load_parameters"> <code class="descname">load_parameters</code>(filename, ctx=None, allow_missing=False, ignore_extra=False, cast_dtype=False, dtype_source='current')<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.load_parameters">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.load_parameters" title="Permalink to this definition">¶</a></dt> <dd>Load parameters from file previously saved by <cite>save_parameters</cite>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>filename (str) – Path to parameter file.</li> <li>ctx (Context or list of Context, default cpu()) – Context(s) to initialize loaded parameters on.</li> <li>allow_missing (bool, default False) – Whether to silently skip loading parameters not represents in the file.</li> <li>ignore_extra (bool, default False) – Whether to silently ignore parameters from the file that are not present in this Block.</li> <li>cast_dtype (bool, default False) – Cast the data type of the NDArray loaded from the checkpoint to the dtype provided by the Parameter if any.</li> <li>dtype_source (str, default 'current') – must be in {‘current’, ‘saved’} Only valid if cast_dtype=True, specify the source of the dtype for casting the parameters</li> </ul> </td> </tr> </tbody> </table> References <a class="reference external" href="/versions/1.5.0/tutorials/gluon/save_load_params.html">Saving and Loading Gluon Models</a> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.load_params"> <code class="descname">load_params</code>(filename, ctx=None, allow_missing=False, ignore_extra=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.load_params">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.load_params" title="Permalink to this definition">¶</a></dt> <dd>[Deprecated] Please use load_parameters. Load parameters from file. <dl class="docutils"> <dt>filename : str</dt> <dd>Path to parameter file.</dd> <dt>ctx : Context or list of Context, default cpu()</dt> <dd>Context(s) to initialize loaded parameters on.</dd> <dt>allow_missing : bool, default False</dt> <dd>Whether to silently skip loading parameters not represents in the file.</dd> <dt>ignore_extra : bool, default False</dt> <dd>Whether to silently ignore parameters from the file that are not present in this Block.</dd> </dl> </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Block.name"> <code class="descname">name</code><a class="headerlink" href="#mxnet.gluon.Block.name" title="Permalink to this definition">¶</a></dt> <dd>Name of this <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a>, without ‘_’ in the end. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.name_scope"> <code class="descname">name_scope</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.name_scope">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.name_scope" title="Permalink to this definition">¶</a></dt> <dd>Returns a name space object managing a child <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> and parameter names. Should be used within a <code class="docutils literal">with</code> statement: <div class="highlight-default"><div class="highlight"><pre>with self.name_scope(): self.dense = nn.Dense(20) </pre></div> </div> Please refer to <a class="reference external" href="/versions/1.5.0/tutorials/gluon/naming.html">naming tutorial</a> for more info on prefix and naming. </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Block.params"> <code class="descname">params</code><a class="headerlink" href="#mxnet.gluon.Block.params" title="Permalink to this definition">¶</a></dt> <dd>Returns this <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a>‘s parameter dictionary (does not include its children’s parameters). </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Block.prefix"> <code class="descname">prefix</code><a class="headerlink" href="#mxnet.gluon.Block.prefix" title="Permalink to this definition">¶</a></dt> <dd>Prefix of this <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a>. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.register_child"> <code class="descname">register_child</code>(block, name=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.register_child">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.register_child" title="Permalink to this definition">¶</a></dt> <dd>Registers block as a child of self. <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a> s assigned to self as attributes will be registered automatically. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.register_forward_hook"> <code class="descname">register_forward_hook</code>(hook)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.register_forward_hook">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.register_forward_hook" title="Permalink to this definition">¶</a></dt> <dd>Registers a forward hook on the block. The hook function is called immediately after <a class="reference internal" href="#mxnet.gluon.Block.forward" title="mxnet.gluon.Block.forward"><code class="xref py py-func docutils literal">forward()</code></a>. It should not modify the input or output. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">hook (callable) – The forward hook function of form <cite>hook(block, input, output) -> None</cite>.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><code class="xref py py-class docutils literal">mxnet.gluon.utils.HookHandle</code></td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.register_forward_pre_hook"> <code class="descname">register_forward_pre_hook</code>(hook)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.register_forward_pre_hook">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.register_forward_pre_hook" title="Permalink to this definition">¶</a></dt> <dd>Registers a forward pre-hook on the block. The hook function is called immediately before <a class="reference internal" href="#mxnet.gluon.Block.forward" title="mxnet.gluon.Block.forward"><code class="xref py py-func docutils literal">forward()</code></a>. It should not modify the input or output. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">hook (callable) – The forward hook function of form <cite>hook(block, input) -> None</cite>.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><code class="xref py py-class docutils literal">mxnet.gluon.utils.HookHandle</code></td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.save_parameters"> <code class="descname">save_parameters</code>(filename)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.save_parameters">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.save_parameters" title="Permalink to this definition">¶</a></dt> <dd>Save parameters to file. Saved parameters can only be loaded with <cite>load_parameters</cite>. Note that this method only saves parameters, not model structure. If you want to save model structures, please use <a class="reference internal" href="#mxnet.gluon.HybridBlock.export" title="mxnet.gluon.HybridBlock.export"><code class="xref py py-meth docutils literal">HybridBlock.export()</code></a>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">filename (str) – Path to file.</td> </tr> </tbody> </table> References <a class="reference external" href="/versions/1.5.0/tutorials/gluon/save_load_params.html">Saving and Loading Gluon Models</a> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.save_params"> <code class="descname">save_params</code>(filename)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.save_params">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.save_params" title="Permalink to this definition">¶</a></dt> <dd>[Deprecated] Please use save_parameters. Note that if you want load from SymbolBlock later, please use export instead. Save parameters to file. <dl class="docutils"> <dt>filename : str</dt> <dd>Path to file.</dd> </dl> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Block.summary"> <code class="descname">summary</code>(*inputs)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#Block.summary">[source]</a><a class="headerlink" href="#mxnet.gluon.Block.summary" title="Permalink to this definition">¶</a></dt> <dd>Print the summary of the model’s output and parameters. The network must have been initialized, and must not have been hybridized. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">inputs (object) – Any input that the model supports. For any tensor in the input, only <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray"><code class="xref py py-class docutils literal">mxnet.ndarray.NDArray</code></a> is supported.</td> </tr> </tbody> </table> </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.Constant"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">Constant</code>(name, value)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Constant">[source]</a><a class="headerlink" href="#mxnet.gluon.Constant" title="Permalink to this definition">¶</a></dt> <dd>A constant parameter for holding immutable tensors. <cite>Constant`s are ignored by `autograd</cite> and <cite>Trainer</cite>, thus their values will not change during training. But you can still update their values manually with the <cite>set_data</cite> method. <cite>Constant</cite> s can be created with either: <div class="highlight-default"><div class="highlight"><pre>const = mx.gluon.Constant('const', [[1,2],[3,4]]) </pre></div> </div> or: <div class="highlight-default"><div class="highlight"><pre>class Block(gluon.Block): def __init__(self, **kwargs): super(Block, self).__init__(**kwargs) self.const = self.params.get_constant('const', [[1,2],[3,4]]) </pre></div> </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>name (str) – Name of the parameter.</li> <li>value (array-like) – Initial value for the constant.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="exception"> <dt id="mxnet.gluon.DeferredInitializationError"> exception <code class="descclassname">mxnet.gluon.</code><code class="descname">DeferredInitializationError</code><a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#DeferredInitializationError">[source]</a><a class="headerlink" href="#mxnet.gluon.DeferredInitializationError" title="Permalink to this definition">¶</a></dt> <dd>Error for unfinished deferred initialization. </dd></dl> <dl class="class"> <dt id="mxnet.gluon.HybridBlock"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">HybridBlock</code>(prefix=None, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock" title="Permalink to this definition">¶</a></dt> <dd><cite>HybridBlock</cite> supports forwarding with both Symbol and NDArray. <cite>HybridBlock</cite> is similar to <cite>Block</cite>, with a few differences: <div class="highlight-default"><div class="highlight"><pre>import mxnet as mx from mxnet.gluon import HybridBlock, nn class Model(HybridBlock): def __init__(self, **kwargs): super(Model, self).__init__(**kwargs) # use name_scope to give child Blocks appropriate names. with self.name_scope(): self.dense0 = nn.Dense(20) self.dense1 = nn.Dense(20) def hybrid_forward(self, F, x): x = F.relu(self.dense0(x)) return F.relu(self.dense1(x)) model = Model() model.initialize(ctx=mx.cpu(0)) model.hybridize() model(mx.nd.zeros((10, 10), ctx=mx.cpu(0))) </pre></div> </div> Forward computation in <a class="reference internal" href="#mxnet.gluon.HybridBlock" title="mxnet.gluon.HybridBlock"><code class="xref py py-class docutils literal">HybridBlock</code></a> must be static to work with <code class="xref py py-class docutils literal">Symbol</code> s, i.e. you cannot call <code class="xref py py-meth docutils literal">NDArray.asnumpy()</code>, <code class="xref py py-attr docutils literal">NDArray.shape</code>, <code class="xref py py-attr docutils literal">NDArray.dtype</code>, <cite>NDArray</cite> indexing (<cite>x[i]</cite>) etc on tensors. Also, you cannot use branching or loop logic that bases on non-constant expressions like random numbers or intermediate results, since they change the graph structure for each iteration. Before activating with <code class="xref py py-meth docutils literal">hybridize()</code>, <a class="reference internal" href="#mxnet.gluon.HybridBlock" title="mxnet.gluon.HybridBlock"><code class="xref py py-class docutils literal">HybridBlock</code></a> works just like normal <a class="reference internal" href="#mxnet.gluon.Block" title="mxnet.gluon.Block"><code class="xref py py-class docutils literal">Block</code></a>. After activation, <a class="reference internal" href="#mxnet.gluon.HybridBlock" title="mxnet.gluon.HybridBlock"><code class="xref py py-class docutils literal">HybridBlock</code></a> will create a symbolic graph representing the forward computation and cache it. On subsequent forwards, the cached graph will be used instead of <a class="reference internal" href="#mxnet.gluon.HybridBlock.hybrid_forward" title="mxnet.gluon.HybridBlock.hybrid_forward"><code class="xref py py-meth docutils literal">hybrid_forward()</code></a>. Please see references for detailed tutorial. References <a class="reference external" href="/versions/1.5.0/tutorials/gluon/hybrid.html">Hybrid - Faster training and easy deployment</a> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.__setattr__"> <code class="descname">__setattr__</code>(name, value)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.__setattr__">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.__setattr__" title="Permalink to this definition">¶</a></dt> <dd>Registers parameters. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.export"> <code class="descname">export</code>(path, epoch=0, remove_amp_cast=True)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.export">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.export" title="Permalink to this definition">¶</a></dt> <dd>Export HybridBlock to json format that can be loaded by <cite>SymbolBlock.imports</cite>, <cite>mxnet.mod.Module</cite> or the C++ interface. <div class="admonition note"> Note When there are only one input, it will have name <cite>data</cite>. When there Are more than one inputs, they will be named as <cite>data0</cite>, <cite>data1</cite>, etc. </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>path (str) – Path to save model. Two files <cite>path-symbol.json</cite> and <cite>path-xxxx.params</cite> will be created, where xxxx is the 4 digits epoch number.</li> <li>epoch (int) – Epoch number of saved model.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.forward"> <code class="descname">forward</code>(x, *args)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.forward">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.forward" title="Permalink to this definition">¶</a></dt> <dd>Defines the forward computation. Arguments can be either <code class="xref py py-class docutils literal">NDArray</code> or <code class="xref py py-class docutils literal">Symbol</code>. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.hybrid_forward"> <code class="descname">hybrid_forward</code>(F, x, *args, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.hybrid_forward">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.hybrid_forward" title="Permalink to this definition">¶</a></dt> <dd>Overrides to construct symbolic graph for this <cite>Block</cite>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>x (<a class="reference internal" href="../symbol/symbol.html#mxnet.symbol.Symbol" title="mxnet.symbol.Symbol">Symbol</a> or <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a>) – The first input tensor.</li> <li>*args (list of Symbol or list of NDArray) – Additional input tensors.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.infer_shape"> <code class="descname">infer_shape</code>(*args)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.infer_shape">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.infer_shape" title="Permalink to this definition">¶</a></dt> <dd>Infers shape of Parameters from inputs. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.HybridBlock.infer_type"> <code class="descname">infer_type</code>(*args)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#HybridBlock.infer_type">[source]</a><a class="headerlink" href="#mxnet.gluon.HybridBlock.infer_type" title="Permalink to this definition">¶</a></dt> <dd>Infers data type of Parameters from inputs. </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.Parameter"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">Parameter</code>(name, grad_req='write', shape=None, dtype=<type 'numpy.float32'>, lr_mult=1.0, wd_mult=1.0, init=None, allow_deferred_init=False, differentiable=True, stype='default', grad_stype='default')<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter" title="Permalink to this definition">¶</a></dt> <dd>A Container holding parameters (weights) of Blocks. <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a> holds a copy of the parameter on each <code class="xref py py-class docutils literal">Context</code> after it is initialized with <code class="docutils literal">Parameter.initialize(...)</code>. If <a class="reference internal" href="#mxnet.gluon.Parameter.grad_req" title="mxnet.gluon.Parameter.grad_req"><code class="xref py py-attr docutils literal">grad_req</code></a> is not <code class="docutils literal">'null'</code>, it will also hold a gradient array on each <code class="xref py py-class docutils literal">Context</code>: <div class="highlight-default"><div class="highlight"><pre>ctx = mx.gpu(0) x = mx.nd.zeros((16, 100), ctx=ctx) w = mx.gluon.Parameter('fc_weight', shape=(64, 100), init=mx.init.Xavier()) b = mx.gluon.Parameter('fc_bias', shape=(64,), init=mx.init.Zero()) w.initialize(ctx=ctx) b.initialize(ctx=ctx) out = mx.nd.FullyConnected(x, w.data(ctx), b.data(ctx), num_hidden=64) </pre></div> </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>name (str) – Name of this parameter.</li> <li>grad_req ({'write', 'add', 'null'}, default 'write') – Specifies how to update gradient to grad arrays. <ul> <li><code class="docutils literal">'write'</code> means everytime gradient is written to grad <code class="xref py py-class docutils literal">NDArray</code>.</li> <li><code class="docutils literal">'add'</code> means everytime gradient is added to the grad <code class="xref py py-class docutils literal">NDArray</code>. You need to manually call <code class="docutils literal">zero_grad()</code> to clear the gradient buffer before each iteration when using this option.</li> <li>‘null’ means gradient is not requested for this parameter. gradient arrays will not be allocated.</li> </ul> </li> <li>shape (int or tuple of int, default None) – Shape of this parameter. By default shape is not specified. Parameter with unknown shape can be used for <code class="xref py py-class docutils literal">Symbol</code> API, but <code class="docutils literal">init</code> will throw an error when using <code class="xref py py-class docutils literal">NDArray</code> API.</li> <li>dtype (numpy.dtype or str, default 'float32') – Data type of this parameter. For example, <code class="docutils literal">numpy.float32</code> or <code class="docutils literal">'float32'</code>.</li> <li>lr_mult (float, default 1.0) – Learning rate multiplier. Learning rate will be multiplied by lr_mult when updating this parameter with optimizer.</li> <li>wd_mult (float, default 1.0) – Weight decay multiplier (L2 regularizer coefficient). Works similar to lr_mult.</li> <li>init (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>, default None) – Initializer of this parameter. Will use the global initializer by default.</li> <li>stype ({'default', 'row_sparse', 'csr'}, defaults to 'default'.) – The storage type of the parameter.</li> <li>grad_stype ({'default', 'row_sparse', 'csr'}, defaults to 'default'.) – The storage type of the parameter’s gradient.</li> </ul> </td> </tr> </tbody> </table> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.grad_req"> <code class="descname">grad_req</code><a class="headerlink" href="#mxnet.gluon.Parameter.grad_req" title="Permalink to this definition">¶</a></dt> <dd>{‘write’, ‘add’, ‘null’} – This can be set before or after initialization. Setting <code class="docutils literal">grad_req</code> to <code class="docutils literal">'null'</code> with <code class="docutils literal">x.grad_req = 'null'</code> saves memory and computation when you don’t need gradient w.r.t x. </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.lr_mult"> <code class="descname">lr_mult</code><a class="headerlink" href="#mxnet.gluon.Parameter.lr_mult" title="Permalink to this definition">¶</a></dt> <dd>float – Local learning rate multiplier for this Parameter. The actual learning rate is calculated with <code class="docutils literal">learning_rate * lr_mult</code>. You can set it with <code class="docutils literal">param.lr_mult = 2.0</code> </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.wd_mult"> <code class="descname">wd_mult</code><a class="headerlink" href="#mxnet.gluon.Parameter.wd_mult" title="Permalink to this definition">¶</a></dt> <dd>float – Local weight decay multiplier for this Parameter. </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.__weakref__"> <code class="descname">__weakref__</code><a class="headerlink" href="#mxnet.gluon.Parameter.__weakref__" title="Permalink to this definition">¶</a></dt> <dd>list of weak references to the object (if defined) </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.cast"> <code class="descname">cast</code>(dtype)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.cast">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.cast" title="Permalink to this definition">¶</a></dt> <dd>Cast data and gradient of this Parameter to a new data type. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">dtype (str or numpy.dtype) – The new data type.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.data"> <code class="descname">data</code>(ctx=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.data">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.data" title="Permalink to this definition">¶</a></dt> <dd>Returns a copy of this parameter on one context. Must have been initialized on this context before. For sparse parameters, use <a class="reference internal" href="#mxnet.gluon.Parameter.row_sparse_data" title="mxnet.gluon.Parameter.row_sparse_data"><code class="xref py py-meth docutils literal">Parameter.row_sparse_data()</code></a> instead. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">ctx (Context) – Desired context.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">NDArray on ctx</td> </tr> </tbody> </table> </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.dtype"> <code class="descname">dtype</code><a class="headerlink" href="#mxnet.gluon.Parameter.dtype" title="Permalink to this definition">¶</a></dt> <dd>The type of the parameter. Setting the dtype value is equivalent to casting the value of the parameter </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.grad"> <code class="descname">grad</code>(ctx=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.grad">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.grad" title="Permalink to this definition">¶</a></dt> <dd>Returns a gradient buffer for this parameter on one context. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">ctx (Context) – Desired context.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.initialize"> <code class="descname">initialize</code>(init=None, ctx=None, default_init=<mxnet.initializer.Uniform object>, force_reinit=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.initialize">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.initialize" title="Permalink to this definition">¶</a></dt> <dd>Initializes parameter and gradient arrays. Only used for <code class="xref py py-class docutils literal">NDArray</code> API. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>init (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – The initializer to use. Overrides <code class="xref py py-meth docutils literal">Parameter.init()</code> and default_init.</li> <li>ctx (Context or list of Context, defaults to <code class="xref py py-meth docutils literal">context.current_context()</code>.) – Initialize Parameter on given context. If ctx is a list of Context, a copy will be made for each context. <div class="admonition note"> Note Copies are independent arrays. User is responsible for keeping their values consistent when updating. Normally <code class="xref py py-class docutils literal">gluon.Trainer</code> does this for you. </div> </li> <li>default_init (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – Default initializer is used when both <code class="xref py py-func docutils literal">init()</code> and <code class="xref py py-meth docutils literal">Parameter.init()</code> are <code class="docutils literal">None</code>.</li> <li>force_reinit (bool, default False) – Whether to force re-initialization if parameter is already initialized.</li> </ul> </td> </tr> </tbody> </table> Examples <div class="highlight-default"><div class="highlight"><pre>>>> weight = mx.gluon.Parameter('weight', shape=(2, 2)) >>> weight.initialize(ctx=mx.cpu(0)) >>> weight.data() [[-0.01068833 0.01729892] [ 0.02042518 -0.01618656]] <NDArray 2x2 @cpu(0)> >>> weight.grad() [[ 0. 0.] [ 0. 0.]] <NDArray 2x2 @cpu(0)> >>> weight.initialize(ctx=[mx.gpu(0), mx.gpu(1)]) >>> weight.data(mx.gpu(0)) [[-0.00873779 -0.02834515] [ 0.05484822 -0.06206018]] <NDArray 2x2 @gpu(0)> >>> weight.data(mx.gpu(1)) [[-0.00873779 -0.02834515] [ 0.05484822 -0.06206018]] <NDArray 2x2 @gpu(1)> </pre></div> </div> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.list_ctx"> <code class="descname">list_ctx</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.list_ctx">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.list_ctx" title="Permalink to this definition">¶</a></dt> <dd>Returns a list of contexts this parameter is initialized on. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.list_data"> <code class="descname">list_data</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.list_data">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.list_data" title="Permalink to this definition">¶</a></dt> <dd>Returns copies of this parameter on all contexts, in the same order as creation. For sparse parameters, use <a class="reference internal" href="#mxnet.gluon.Parameter.list_row_sparse_data" title="mxnet.gluon.Parameter.list_row_sparse_data"><code class="xref py py-meth docutils literal">Parameter.list_row_sparse_data()</code></a> instead. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-even field"><th class="field-name">Return type:</th><td class="field-body">list of NDArrays</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.list_grad"> <code class="descname">list_grad</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.list_grad">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.list_grad" title="Permalink to this definition">¶</a></dt> <dd>Returns gradient buffers on all contexts, in the same order as <code class="xref py py-meth docutils literal">values()</code>. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.list_row_sparse_data"> <code class="descname">list_row_sparse_data</code>(row_id)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.list_row_sparse_data">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.list_row_sparse_data" title="Permalink to this definition">¶</a></dt> <dd>Returns copies of the ‘row_sparse’ parameter on all contexts, in the same order as creation. The copy only retains rows whose ids occur in provided row ids. The parameter must have been initialized before. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">row_id (<a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a>) – Row ids to retain for the ‘row_sparse’ parameter.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">list of NDArrays</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.reset_ctx"> <code class="descname">reset_ctx</code>(ctx)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.reset_ctx">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.reset_ctx" title="Permalink to this definition">¶</a></dt> <dd>Re-assign Parameter to other contexts. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">ctx (Context or list of Context, default <code class="docutils literal">context.current_context()</code>.) – Assign Parameter to given context. If ctx is a list of Context, a copy will be made for each context.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.row_sparse_data"> <code class="descname">row_sparse_data</code>(row_id)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.row_sparse_data">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.row_sparse_data" title="Permalink to this definition">¶</a></dt> <dd>Returns a copy of the ‘row_sparse’ parameter on the same context as row_id’s. The copy only retains rows whose ids occur in provided row ids. The parameter must have been initialized on this context before. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">row_id (<a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a>) – Row ids to retain for the ‘row_sparse’ parameter.</td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body"></td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">NDArray on row_id’s context</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.set_data"> <code class="descname">set_data</code>(data)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.set_data">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.set_data" title="Permalink to this definition">¶</a></dt> <dd>Sets this parameter’s value on all contexts. </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.Parameter.shape"> <code class="descname">shape</code><a class="headerlink" href="#mxnet.gluon.Parameter.shape" title="Permalink to this definition">¶</a></dt> <dd>The shape of the parameter. By default, an unknown dimension size is 0. However, when the NumPy semantic is turned on, unknown dimension size is -1. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.var"> <code class="descname">var</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.var">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.var" title="Permalink to this definition">¶</a></dt> <dd>Returns a symbol representing this parameter. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Parameter.zero_grad"> <code class="descname">zero_grad</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#Parameter.zero_grad">[source]</a><a class="headerlink" href="#mxnet.gluon.Parameter.zero_grad" title="Permalink to this definition">¶</a></dt> <dd>Sets gradient buffer on all contexts to 0. No action is taken if parameter is uninitialized or doesn’t require gradient. </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.ParameterDict"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">ParameterDict</code>(prefix='', shared=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict" title="Permalink to this definition">¶</a></dt> <dd>A dictionary managing a set of parameters. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>prefix (str, default <code class="docutils literal">''</code>) – The prefix to be prepended to all Parameters’ names created by this dict.</li> <li>shared (<a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict">ParameterDict</a> or None) – If not <code class="docutils literal">None</code>, when this dict’s <a class="reference internal" href="#mxnet.gluon.ParameterDict.get" title="mxnet.gluon.ParameterDict.get"><code class="xref py py-meth docutils literal">get()</code></a> method creates a new parameter, will first try to retrieve it from “shared” dict. Usually used for sharing parameters with another Block.</li> </ul> </td> </tr> </tbody> </table> <dl class="attribute"> <dt id="mxnet.gluon.ParameterDict.__weakref__"> <code class="descname">__weakref__</code><a class="headerlink" href="#mxnet.gluon.ParameterDict.__weakref__" title="Permalink to this definition">¶</a></dt> <dd>list of weak references to the object (if defined) </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.get"> <code class="descname">get</code>(name, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.get">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.get" title="Permalink to this definition">¶</a></dt> <dd>Retrieves a <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a> with name <code class="docutils literal">self.prefix+name</code>. If not found, <a class="reference internal" href="#mxnet.gluon.ParameterDict.get" title="mxnet.gluon.ParameterDict.get"><code class="xref py py-func docutils literal">get()</code></a> will first try to retrieve it from “shared” dict. If still not found, <a class="reference internal" href="#mxnet.gluon.ParameterDict.get" title="mxnet.gluon.ParameterDict.get"><code class="xref py py-func docutils literal">get()</code></a> will create a new <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a> with key-word arguments and insert it to self. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>name (str) – Name of the desired Parameter. It will be prepended with this dictionary’s prefix.</li> <li>**kwargs (dict) – The rest of key-word arguments for the created <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a>.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">The created or retrieved <a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter"><code class="xref py py-class docutils literal">Parameter</code></a>. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><a class="reference internal" href="#mxnet.gluon.Parameter" title="mxnet.gluon.Parameter">Parameter</a> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.get_constant"> <code class="descname">get_constant</code>(name, value=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.get_constant">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.get_constant" title="Permalink to this definition">¶</a></dt> <dd>Retrieves a <a class="reference internal" href="#mxnet.gluon.Constant" title="mxnet.gluon.Constant"><code class="xref py py-class docutils literal">Constant</code></a> with name <code class="docutils literal">self.prefix+name</code>. If not found, <a class="reference internal" href="#mxnet.gluon.ParameterDict.get" title="mxnet.gluon.ParameterDict.get"><code class="xref py py-func docutils literal">get()</code></a> will first try to retrieve it from “shared” dict. If still not found, <a class="reference internal" href="#mxnet.gluon.ParameterDict.get" title="mxnet.gluon.ParameterDict.get"><code class="xref py py-func docutils literal">get()</code></a> will create a new <a class="reference internal" href="#mxnet.gluon.Constant" title="mxnet.gluon.Constant"><code class="xref py py-class docutils literal">Constant</code></a> with key-word arguments and insert it to self. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>name (str) – Name of the desired Constant. It will be prepended with this dictionary’s prefix.</li> <li>value (array-like) – Initial value of constant.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">The created or retrieved <a class="reference internal" href="#mxnet.gluon.Constant" title="mxnet.gluon.Constant"><code class="xref py py-class docutils literal">Constant</code></a>. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><a class="reference internal" href="#mxnet.gluon.Constant" title="mxnet.gluon.Constant"><code class="xref py py-class docutils literal">Constant</code></a> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.initialize"> <code class="descname">initialize</code>(init=<mxnet.initializer.Uniform object>, ctx=None, verbose=False, force_reinit=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.initialize">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.initialize" title="Permalink to this definition">¶</a></dt> <dd>Initializes all Parameters managed by this dictionary to be used for <code class="xref py py-class docutils literal">NDArray</code> API. It has no effect when using <code class="xref py py-class docutils literal">Symbol</code> API. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>init (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – Global default Initializer to be used when <code class="xref py py-meth docutils literal">Parameter.init()</code> is <code class="docutils literal">None</code>. Otherwise, <code class="xref py py-meth docutils literal">Parameter.init()</code> takes precedence.</li> <li>ctx (Context or list of Context) – Keeps a copy of Parameters on one or many context(s).</li> <li>verbose (bool, default False) – Whether to verbosely print out details on initialization.</li> <li>force_reinit (bool, default False) – Whether to force re-initialization if parameter is already initialized.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.load"> <code class="descname">load</code>(filename, ctx=None, allow_missing=False, ignore_extra=False, restore_prefix='', cast_dtype=False, dtype_source='current')<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.load">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.load" title="Permalink to this definition">¶</a></dt> <dd>Load parameters from file. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>filename (str) – Path to parameter file.</li> <li>ctx (Context or list of Context) – Context(s) initialize loaded parameters on.</li> <li>allow_missing (bool, default False) – Whether to silently skip loading parameters not represents in the file.</li> <li>ignore_extra (bool, default False) – Whether to silently ignore parameters from the file that are not present in this ParameterDict.</li> <li>restore_prefix (str, default '') – prepend prefix to names of stored parameters before loading.</li> <li>cast_dtype (bool, default False) – Cast the data type of the parameter</li> <li>dtype_source (str, default 'current') – must be in {‘current’, ‘saved’} Only valid if cast_dtype=True, specify the source of the dtype for casting the parameters</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="attribute"> <dt id="mxnet.gluon.ParameterDict.prefix"> <code class="descname">prefix</code><a class="headerlink" href="#mxnet.gluon.ParameterDict.prefix" title="Permalink to this definition">¶</a></dt> <dd>Prefix of this dict. It will be prepended to <code class="xref py py-class docutils literal">Parameter`s' name created with :py:func:`get</code>. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.reset_ctx"> <code class="descname">reset_ctx</code>(ctx)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.reset_ctx">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.reset_ctx" title="Permalink to this definition">¶</a></dt> <dd>Re-assign all Parameters to other contexts. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">ctx (Context or list of Context, default <code class="xref py py-meth docutils literal">context.current_context()</code>.) – Assign Parameter to given context. If ctx is a list of Context, a copy will be made for each context.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.save"> <code class="descname">save</code>(filename, strip_prefix='')<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.save">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.save" title="Permalink to this definition">¶</a></dt> <dd>Save parameters to file. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>filename (str) – Path to parameter file.</li> <li>strip_prefix (str, default '') – Strip prefix from parameter names before saving.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.setattr"> <code class="descname">setattr</code>(name, value)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.setattr">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.setattr" title="Permalink to this definition">¶</a></dt> <dd>Set an attribute to a new value for all Parameters. For example, set grad_req to null if you don’t need gradient w.r.t a model’s Parameters: <div class="highlight-default"><div class="highlight"><pre>model.collect_params().setattr('grad_req', 'null') </pre></div> </div> or change the learning rate multiplier: <div class="highlight-default"><div class="highlight"><pre>model.collect_params().setattr('lr_mult', 0.5) </pre></div> </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>name (str) – Name of the attribute.</li> <li>value (valid type for attribute name) – The new value for the attribute.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.update"> <code class="descname">update</code>(other)<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.update">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.update" title="Permalink to this definition">¶</a></dt> <dd>Copies all Parameters in <code class="docutils literal">other</code> to self. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.ParameterDict.zero_grad"> <code class="descname">zero_grad</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/parameter.html#ParameterDict.zero_grad">[source]</a><a class="headerlink" href="#mxnet.gluon.ParameterDict.zero_grad" title="Permalink to this definition">¶</a></dt> <dd>Sets all Parameters’ gradient buffer to 0. </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.SymbolBlock"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">SymbolBlock</code>(outputs, inputs, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#SymbolBlock">[source]</a><a class="headerlink" href="#mxnet.gluon.SymbolBlock" title="Permalink to this definition">¶</a></dt> <dd>Construct block from symbol. This is useful for using pre-trained models as feature extractors. For example, you may want to extract the output from fc2 layer in AlexNet. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>outputs (<a class="reference internal" href="../symbol/symbol.html#mxnet.symbol.Symbol" title="mxnet.symbol.Symbol">Symbol</a> or list of Symbol) – The desired output for SymbolBlock.</li> <li>inputs (<a class="reference internal" href="../symbol/symbol.html#mxnet.symbol.Symbol" title="mxnet.symbol.Symbol">Symbol</a> or list of Symbol) – The Variables in output’s argument that should be used as inputs.</li> <li>params (<a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict">ParameterDict</a>) – Parameter dictionary for arguments and auxililary states of outputs that are not inputs.</li> </ul> </td> </tr> </tbody> </table> Examples <div class="highlight-default"><div class="highlight"><pre>>>> # To extract the feature from fc1 and fc2 layers of AlexNet: >>> alexnet = gluon.model_zoo.vision.alexnet(pretrained=True, ctx=mx.cpu(), prefix='model_') >>> inputs = mx.sym.var('data') >>> out = alexnet(inputs) >>> internals = out.get_internals() >>> print(internals.list_outputs()) ['data', ..., 'model_dense0_relu_fwd_output', ..., 'model_dense1_relu_fwd_output', ...] >>> outputs = [internals['model_dense0_relu_fwd_output'], internals['model_dense1_relu_fwd_output']] >>> # Create SymbolBlock that shares parameters with alexnet >>> feat_model = gluon.SymbolBlock(outputs, inputs, params=alexnet.collect_params()) >>> x = mx.nd.random.normal(shape=(16, 3, 224, 224)) >>> print(feat_model(x)) </pre></div> </div> <dl class="staticmethod"> <dt id="mxnet.gluon.SymbolBlock.imports"> static <code class="descname">imports</code>(symbol_file, input_names, param_file=None, ctx=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/block.html#SymbolBlock.imports">[source]</a><a class="headerlink" href="#mxnet.gluon.SymbolBlock.imports" title="Permalink to this definition">¶</a></dt> <dd>Import model previously saved by <cite>HybridBlock.export</cite> or <cite>Module.save_checkpoint</cite> as a SymbolBlock for use in Gluon. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>symbol_file (str) – Path to symbol file.</li> <li>input_names (list of str) – List of input variable names</li> <li>param_file (str, optional) – Path to parameter file.</li> <li>ctx (Context, default None) – The context to initialize SymbolBlock on.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">SymbolBlock loaded from symbol and parameter files. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><a class="reference internal" href="#mxnet.gluon.SymbolBlock" title="mxnet.gluon.SymbolBlock">SymbolBlock</a> </td> </tr> </tbody> </table> Examples <div class="highlight-default"><div class="highlight"><pre>>>> net1 = gluon.model_zoo.vision.resnet18_v1( ... prefix='resnet', pretrained=True) >>> net1.hybridize() >>> x = mx.nd.random.normal(shape=(1, 3, 32, 32)) >>> out1 = net1(x) >>> net1.export('net1', epoch=1) >>> >>> net2 = gluon.SymbolBlock.imports( ... 'net1-symbol.json', ['data'], 'net1-0001.params') >>> out2 = net2(x) </pre></div> </div> </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.Trainer"> class <code class="descclassname">mxnet.gluon.</code><code class="descname">Trainer</code>(params, optimizer, optimizer_params=None, kvstore='device', compression_params=None, update_on_kvstore=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer" title="Permalink to this definition">¶</a></dt> <dd>Applies an <cite>Optimizer</cite> on a set of Parameters. Trainer should be used together with <cite>autograd</cite>. <div class="admonition note"> Note For the following cases, updates will always happen on kvstore, i.e., you cannot set update_on_kvstore=False. <ul class="last simple"> <li>dist kvstore with sparse weights or sparse gradients</li> <li>dist async kvstore</li> <li><cite>optimizer.lr_scheduler</cite> is not None</li> </ul> </div> <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>params (<a class="reference internal" href="#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict">ParameterDict</a>) – The set of parameters to optimize.</li> <li>optimizer (str or <a class="reference internal" href="../optimization/optimization.html#mxnet.optimizer.Optimizer" title="mxnet.optimizer.Optimizer">Optimizer</a>) – The optimizer to use. See <a class="reference external" href="/api/python/optimization/optimization.html#the-mxnet-optimizer-package">help</a> on Optimizer for a list of available optimizers.</li> <li>optimizer_params (dict) – Key-word arguments to be passed to optimizer constructor. For example, <cite>{‘learning_rate’: 0.1}</cite>. All optimizers accept learning_rate, wd (weight decay), clip_gradient, and lr_scheduler. See each optimizer’s constructor for a list of additional supported arguments.</li> <li>kvstore (str or <a class="reference internal" href="../kvstore/kvstore.html#mxnet.kvstore.KVStore" title="mxnet.kvstore.KVStore">KVStore</a>) – kvstore type for multi-gpu and distributed training. See help on <a class="reference internal" href="../kvstore/kvstore.html#mxnet.kvstore.create" title="mxnet.kvstore.create"><code class="xref any py py-func docutils literal">mxnet.kvstore.create</code></a> for more information.</li> <li>compression_params (dict) – Specifies type of gradient compression and additional arguments depending on the type of compression being used. For example, 2bit compression requires a threshold. Arguments would then be {‘type’:‘2bit’, ‘threshold’:0.5} See mxnet.KVStore.set_gradient_compression method for more details on gradient compression.</li> <li>update_on_kvstore (bool, default None) – Whether to perform parameter updates on kvstore. If None, then trainer will choose the more suitable option depending on the type of kvstore. If the <cite>update_on_kvstore</cite> argument is provided, environment variable <cite>MXNET_UPDATE_ON_KVSTORE</cite> will be ignored.</li> <li>Properties – </li> <li>---------- – </li> <li>learning_rate (float) – The current learning rate of the optimizer. Given an Optimizer object optimizer, its learning rate can be accessed as optimizer.learning_rate.</li> </ul> </td> </tr> </tbody> </table> <dl class="attribute"> <dt id="mxnet.gluon.Trainer.__weakref__"> <code class="descname">__weakref__</code><a class="headerlink" href="#mxnet.gluon.Trainer.__weakref__" title="Permalink to this definition">¶</a></dt> <dd>list of weak references to the object (if defined) </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.allreduce_grads"> <code class="descname">allreduce_grads</code>()<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.allreduce_grads">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.allreduce_grads" title="Permalink to this definition">¶</a></dt> <dd>For each parameter, reduce the gradients from different contexts. Should be called after <cite>autograd.backward()</cite>, outside of <cite>record()</cite> scope, and before <cite>trainer.update()</cite>. For normal parameter updates, <cite>step()</cite> should be used, which internally calls <cite>allreduce_grads()</cite> and then <cite>update()</cite>. However, if you need to get the reduced gradients to perform certain transformation, such as in gradient clipping, then you may want to manually call <cite>allreduce_grads()</cite> and <cite>update()</cite> separately. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.load_states"> <code class="descname">load_states</code>(fname)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.load_states">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.load_states" title="Permalink to this definition">¶</a></dt> <dd>Loads trainer states (e.g. optimizer, momentum) from a file. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">fname (str) – Path to input states file.</td> </tr> </tbody> </table> <div class="admonition note"> Note <cite>optimizer.param_dict</cite>, which contains Parameter information (such as <cite>lr_mult</cite> and <cite>wd_mult</cite>) will not be loaded from the file, but rather set based on current Trainer’s parameters. </div> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.save_states"> <code class="descname">save_states</code>(fname)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.save_states">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.save_states" title="Permalink to this definition">¶</a></dt> <dd>Saves trainer states (e.g. optimizer, momentum) to a file. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">fname (str) – Path to output states file.</td> </tr> </tbody> </table> <div class="admonition note"> Note <cite>optimizer.param_dict</cite>, which contains Parameter information (such as <cite>lr_mult</cite> and <cite>wd_mult</cite>) will not be saved. </div> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.set_learning_rate"> <code class="descname">set_learning_rate</code>(lr)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.set_learning_rate">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.set_learning_rate" title="Permalink to this definition">¶</a></dt> <dd>Sets a new learning rate of the optimizer. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body">lr (float) – The new learning rate of the optimizer.</td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.step"> <code class="descname">step</code>(batch_size, ignore_stale_grad=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.step">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.step" title="Permalink to this definition">¶</a></dt> <dd>Makes one step of parameter update. Should be called after <cite>autograd.backward()</cite> and outside of <cite>record()</cite> scope. For normal parameter updates, <cite>step()</cite> should be used, which internally calls <cite>allreduce_grads()</cite> and then <cite>update()</cite>. However, if you need to get the reduced gradients to perform certain transformation, such as in gradient clipping, then you may want to manually call <cite>allreduce_grads()</cite> and <cite>update()</cite> separately. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>batch_size (int) – Batch size of data processed. Gradient will be normalized by <cite>1/batch_size</cite>. Set this to 1 if you normalized loss manually with <cite>loss = mean(loss)</cite>.</li> <li>ignore_stale_grad (bool, optional, default=False) – If true, ignores Parameters with stale gradient (gradient that has not been updated by <cite>backward</cite> after last step) and skip update.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="method"> <dt id="mxnet.gluon.Trainer.update"> <code class="descname">update</code>(batch_size, ignore_stale_grad=False)<a class="reference internal" href="../../../_modules/mxnet/gluon/trainer.html#Trainer.update">[source]</a><a class="headerlink" href="#mxnet.gluon.Trainer.update" title="Permalink to this definition">¶</a></dt> <dd>Makes one step of parameter update. Should be called after <cite>autograd.backward()</cite> and outside of <cite>record()</cite> scope, and after <cite>trainer.update()</cite>. For normal parameter updates, <cite>step()</cite> should be used, which internally calls <cite>allreduce_grads()</cite> and then <cite>update()</cite>. However, if you need to get the reduced gradients to perform certain transformation, such as in gradient clipping, then you may want to manually call <cite>allreduce_grads()</cite> and <cite>update()</cite> separately. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>batch_size (int) – Batch size of data processed. Gradient will be normalized by <cite>1/batch_size</cite>. Set this to 1 if you normalized loss manually with <cite>loss = mean(loss)</cite>.</li> <li>ignore_stale_grad (bool, optional, default=False) – If true, ignores Parameters with stale gradient (gradient that has not been updated by <cite>backward</cite> after last step) and skip update.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Sequential"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Sequential</code>(prefix=None, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Sequential">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Sequential" title="Permalink to this definition">¶</a></dt> <dd>Stacks Blocks sequentially. Example: <div class="highlight-default"><div class="highlight"><pre>net = nn.Sequential() # use net's name_scope to give child Blocks appropriate names. with net.name_scope(): net.add(nn.Dense(10, activation='relu')) net.add(nn.Dense(20)) </pre></div> </div> <dl class="method"> <dt id="mxnet.gluon.nn.Sequential.add"> <code class="descname">add</code>(*blocks)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Sequential.add">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Sequential.add" title="Permalink to this definition">¶</a></dt> <dd>Adds block on top of the stack. </dd></dl> <dl class="method"> <dt id="mxnet.gluon.nn.Sequential.hybridize"> <code class="descname">hybridize</code>(active=True, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Sequential.hybridize">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Sequential.hybridize" title="Permalink to this definition">¶</a></dt> <dd>Activates or deactivates <cite>HybridBlock</cite> s recursively. Has no effect on non-hybrid children. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first last simple"> <li>active (bool, default True) – Whether to turn hybrid on or off.</li> <li>**kwargs (string) – Additional flags for hybridized operator.</li> </ul> </td> </tr> </tbody> </table> </dd></dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.HybridSequential"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">HybridSequential</code>(prefix=None, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#HybridSequential">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.HybridSequential" title="Permalink to this definition">¶</a></dt> <dd>Stacks HybridBlocks sequentially. Example: <div class="highlight-default"><div class="highlight"><pre>net = nn.HybridSequential() # use net's name_scope to give child Blocks appropriate names. with net.name_scope(): net.add(nn.Dense(10, activation='relu')) net.add(nn.Dense(20)) net.hybridize() </pre></div> </div> <dl class="method"> <dt id="mxnet.gluon.nn.HybridSequential.add"> <code class="descname">add</code>(*blocks)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#HybridSequential.add">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.HybridSequential.add" title="Permalink to this definition">¶</a></dt> <dd>Adds block on top of the stack. </dd></dl> </dd></dl> Parallelization utility optimizer. <dl class="function"> <dt id="mxnet.gluon.utils.split_data"> <code class="descclassname">mxnet.gluon.utils.</code><code class="descname">split_data</code>(data, num_slice, batch_axis=0, even_split=True)<a class="reference internal" href="../../../_modules/mxnet/gluon/utils.html#split_data">[source]</a><a class="headerlink" href="#mxnet.gluon.utils.split_data" title="Permalink to this definition">¶</a></dt> <dd>Splits an NDArray into <cite>num_slice</cite> slices along <cite>batch_axis</cite>. Usually used for data parallelism where each slices is sent to one device (i.e. GPU). <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>data (<a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a>) – A batch of data.</li> <li>num_slice (int) – Number of desired slices.</li> <li>batch_axis (int, default 0) – The axis along which to slice.</li> <li>even_split (bool, default True) – Whether to force all slices to have the same number of elements. If <cite>True</cite>, an error will be raised when <cite>num_slice</cite> does not evenly divide <cite>data.shape[batch_axis]</cite>.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">Return value is a list even if <cite>num_slice</cite> is 1. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">list of NDArray </td> </tr> </tbody> </table> </dd></dl> <dl class="function"> <dt id="mxnet.gluon.utils.split_and_load"> <code class="descclassname">mxnet.gluon.utils.</code><code class="descname">split_and_load</code>(data, ctx_list, batch_axis=0, even_split=True)<a class="reference internal" href="../../../_modules/mxnet/gluon/utils.html#split_and_load">[source]</a><a class="headerlink" href="#mxnet.gluon.utils.split_and_load" title="Permalink to this definition">¶</a></dt> <dd>Splits an NDArray into <cite>len(ctx_list)</cite> slices along <cite>batch_axis</cite> and loads each slice to one context in <cite>ctx_list</cite>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>data (<a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a>) – A batch of data.</li> <li>ctx_list (list of Context) – A list of Contexts.</li> <li>batch_axis (int, default 0) – The axis along which to slice.</li> <li>even_split (bool, default True) – Whether to force all slices to have the same number of elements.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">Each corresponds to a context in <cite>ctx_list</cite>. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">list of NDArray </td> </tr> </tbody> </table> </dd></dl> <dl class="function"> <dt id="mxnet.gluon.utils.clip_global_norm"> <code class="descclassname">mxnet.gluon.utils.</code><code class="descname">clip_global_norm</code>(arrays, max_norm, check_isfinite=True)<a class="reference internal" href="../../../_modules/mxnet/gluon/utils.html#clip_global_norm">[source]</a><a class="headerlink" href="#mxnet.gluon.utils.clip_global_norm" title="Permalink to this definition">¶</a></dt> <dd>Rescales NDArrays so that the sum of their 2-norm is smaller than <cite>max_norm</cite>. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>arrays (list of NDArray) – </li> <li>max_norm (float) – </li> <li>check_isfinite (bool, default True) – If True, check that the total_norm is finite (not nan or inf). This requires a blocking .asscalar() call.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">Total norm. Return type is NDArray of shape (1,) if check_isfinite is False. Otherwise a float is returned. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body"><a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.NDArray" title="mxnet.ndarray.NDArray">NDArray</a> or float </td> </tr> </tbody> </table> </dd></dl> <dl class="function"> <dt id="mxnet.gluon.utils.check_sha1"> <code class="descclassname">mxnet.gluon.utils.</code><code class="descname">check_sha1</code>(filename, sha1_hash)<a class="reference internal" href="../../../_modules/mxnet/gluon/utils.html#check_sha1">[source]</a><a class="headerlink" href="#mxnet.gluon.utils.check_sha1" title="Permalink to this definition">¶</a></dt> <dd>Check whether the sha1 hash of the file content matches the expected hash. <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>filename (str) – Path to the file.</li> <li>sha1_hash (str) – Expected sha1 hash in hexadecimal digits.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">Whether the file content matches the expected hash. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">bool </td> </tr> </tbody> </table> </dd></dl> <dl class="function"> <dt id="mxnet.gluon.utils.download"> <code class="descclassname">mxnet.gluon.utils.</code><code class="descname">download</code>(url, path=None, overwrite=False, sha1_hash=None, retries=5, verify_ssl=True)<a class="reference internal" href="../../../_modules/mxnet/gluon/utils.html#download">[source]</a><a class="headerlink" href="#mxnet.gluon.utils.download" title="Permalink to this definition">¶</a></dt> <dd>Download an given URL <table class="docutils field-list" frame="void" rules="none"> <col class="field-name"/> <col class="field-body"/> <tbody valign="top"> <tr class="field-odd field"><th class="field-name">Parameters:</th><td class="field-body"><ul class="first simple"> <li>url (str) – URL to download</li> <li>path (str, optional) – Destination path to store downloaded file. By default stores to the current directory with same name as in url.</li> <li>overwrite (bool, optional) – Whether to overwrite destination file if already exists.</li> <li>sha1_hash (str, optional) – Expected sha1 hash in hexadecimal digits. Will ignore existing file when hash is specified but doesn’t match.</li> <li>retries (integer, default 5) – The number of times to attempt the download in case of failure or non 200 return codes</li> <li>verify_ssl (bool, default True) – Verify SSL certificates.</li> </ul> </td> </tr> <tr class="field-even field"><th class="field-name">Returns:</th><td class="field-body">The file path of the downloaded file. </td> </tr> <tr class="field-odd field"><th class="field-name">Return type:</th><td class="field-body">str </td> </tr> </tbody> </table> </dd></dl> <script>auto_index("api-reference");</script></div> </div> </div> </div> <div aria-label="main navigation" class="sphinxsidebar rightsidebar" role="navigation"> <div class="sphinxsidebarwrapper"> <h3><a href="../../../index.html">Table Of Contents</a></h3> <ul> <li><a class="reference internal" href="#">Gluon Package</a><ul> <li><a class="reference internal" href="#overview">Overview</a></li> <li><a class="reference internal" href="#contents">Contents</a></li> <li><a class="reference internal" href="#parameter">Parameter</a></li> <li><a class="reference internal" href="#containers">Containers</a></li> <li><a class="reference internal" href="#trainer">Trainer</a></li> <li><a class="reference internal" href="#utilities">Utilities</a></li> <li><a class="reference internal" href="#api-reference">API Reference</a></li> </ul> </li> </ul> </div> </div> </div><div class="footer"> <div class="section-disclaimer"> <div class="container"> <div> <img height="60" src="https://raw.githubusercontent.com/dmlc/web-data/master/mxnet/image/apache_incubator_logo.png"/> Apache MXNet is an effort undergoing incubation at The Apache Software Foundation (ASF), sponsored by the Apache Incubator. Incubation is required of all newly accepted projects until a further review indicates that the infrastructure, communications, and decision making process have stabilized in a manner consistent with other successful ASF projects. While incubation status is not necessarily a reflection of the completeness or stability of the code, it does indicate that the project has yet to be fully endorsed by the ASF. "Copyright © 2017-2018, The Apache Software Foundation Apache MXNet, MXNet, Apache, the Apache feather, and the Apache MXNet project logo are either registered trademarks or trademarks of the Apache Software Foundation." </div> </div> </div> </div>  </div> <script crossorigin="anonymous" integrity="sha384-0mSbJDEHialfmuBBQP6A4Qrprq5OVfW37PRR3j5ELqxss1yVqOtnepnHVP9aJ7xS" src="https://maxcdn.bootstrapcdn.com/bootstrap/3.3.6/js/bootstrap.min.js"></script> <script src="../../../_static/js/sidebar.js" type="text/javascript"></script> <script src="../../../_static/js/search.js" type="text/javascript"></script> <script src="../../../_static/js/navbar.js" type="text/javascript"></script> <script src="../../../_static/js/clipboard.min.js" type="text/javascript"></script> <script src="../../../_static/js/copycode.js" type="text/javascript"></script> <script src="../../../_static/js/page.js" type="text/javascript"></script> <script src="../../../_static/js/docversion.js" type="text/javascript"></script> <script type="text/javascript"> $('body').ready(function () { $('body').css('visibility', 'visible'); }); </script> </body> </html>

versions/1.5.0/api/python/gluon/gluon.html (1,775 lines of code) (raw):