versions/1.4.1/api/python/gluon/nn.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 Neural Network Layers" 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 Neural Network Layers" property="og:description"/> <title>Gluon Neural Network Layers — 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/1.4.1/searchindex.js"); Search.init();}); </script>          <link href="../../../genindex.html" rel="index" title="Index"> <link href="../../../search.html" rel="search" title="Search"/> <link href="gluon.html" rel="up" title="Gluon Package"/> <link href="rnn.html" rel="next" title="Gluon Recurrent Neural Network API"/> <link href="gluon.html" rel="prev" title="Gluon Package"/> <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/1.4.1/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/1.4.1/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/1.4.1/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/clojure/index.html">Clojure</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/java/index.html">Java</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/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/1.4.1/faq/index.html">FAQ</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/tutorials/index.html">Tutorials</a> <li><a class="main-nav-link" href="https://github.com/apache/incubator-mxnet/tree/1.4.1/example">Examples</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/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/1.4.1/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/1.4.1">Github</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/community/contribute.html">Contribute</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/community/ecosystem.html">Ecosystem</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/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">1.4.1</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/1.4.1/install/index.html">Install</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/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/1.4.1/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/1.4.1/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/clojure/index.html">Clojure</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/java/index.html">Java</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/1.4.1/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/1.4.1/faq/index.html" tabindex="-1">FAQ</a></li> <li><a href="/versions/1.4.1/tutorials/index.html" tabindex="-1">Tutorials</a></li> <li><a href="https://github.com/apache/incubator-mxnet/tree/1.4.1/example" tabindex="-1">Examples</a></li> <li><a href="/versions/1.4.1/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/1.4.1/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/1.4.1" tabindex="-1">Github</a></li> <li><a href="/versions/1.4.1/community/contribute.html" tabindex="-1">Contribute</a></li> <li><a href="/versions/1.4.1/community/ecosystem.html" tabindex="-1">Ecosystem</a></li> <li><a href="/versions/1.4.1/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">1.4.1</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 class="current"> <li class="toctree-l1 current"><a class="reference internal" href="../../index.html">MXNet APIs</a><ul class="current"> <li class="toctree-l2"><a class="reference internal" href="../../c++/index.html">MXNet - C++ API</a></li> <li class="toctree-l2"><a class="reference internal" href="../../clojure/index.html">MXNet - Clojure API</a></li> <li class="toctree-l2"><a class="reference internal" href="../../julia/index.html">MXNet - Julia API</a></li> <li class="toctree-l2"><a class="reference internal" href="../../perl/index.html">MXNet - Perl API</a></li> <li class="toctree-l2 current"><a class="reference internal" href="../index.html">MXNet - Python API</a><ul class="current"> <li class="toctree-l3"><a class="reference internal" href="../index.html#autograd-api">Autograd API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#callback-api">Callback API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#contrib-package">Contrib Package</a></li> <li class="toctree-l3 current"><a class="reference internal" href="../index.html#gluon-api">Gluon API</a><ul class="current"> <li class="toctree-l4 current"><a class="reference internal" href="gluon.html">Gluon Package</a></li> <li class="toctree-l4 current"><a class="current reference internal" href="#">Gluon Neural Network Layers</a></li> <li class="toctree-l4"><a class="reference internal" href="rnn.html">Gluon Recurrent Neural Network API</a></li> <li class="toctree-l4"><a class="reference internal" href="loss.html">Gluon Loss API</a></li> <li class="toctree-l4"><a class="reference internal" href="data.html">Gluon Data API</a></li> <li class="toctree-l4"><a class="reference internal" href="model_zoo.html">Gluon Model Zoo</a></li> <li class="toctree-l4"><a class="reference internal" href="contrib.html">Gluon Contrib API</a></li> </ul> </li> <li class="toctree-l3"><a class="reference internal" href="../index.html#image-api">Image API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#io-api">IO API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#kv-store-api">KV Store API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#metric-api">Metric API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#module-api">Module API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#ndarray-api">NDArray API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#optimization-api">Optimization API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#profiler-api">Profiler API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#run-time-compilation-api">Run-Time Compilation API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#symbol-api">Symbol API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#symbol-in-pictures-api">Symbol in Pictures API</a></li> <li class="toctree-l3"><a class="reference internal" href="../index.html#tools">Tools</a></li> </ul> </li> <li class="toctree-l2"><a class="reference internal" href="../../r/index.html">MXNet - R API</a></li> <li class="toctree-l2"><a class="reference internal" href="../../scala/index.html">MXNet - Scala API</a></li> </ul> </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-neural-network-layers"> <h1>Gluon Neural Network Layers<a class="headerlink" href="#gluon-neural-network-layers" title="Permalink to this headline">¶</a></h1> <div class="section" id="overview"> <h2>Overview<a class="headerlink" href="#overview" title="Permalink to this headline">¶</a></h2> This document lists the neural network blocks in Gluon: </div> <div class="section" id="basic-layers"> <h2>Basic Layers<a class="headerlink" href="#basic-layers" 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.nn.Dense" title="mxnet.gluon.nn.Dense"><code class="xref py py-obj docutils literal">Dense</code></a></td> <td>Just your regular densely-connected NN layer.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Dropout" title="mxnet.gluon.nn.Dropout"><code class="xref py py-obj docutils literal">Dropout</code></a></td> <td>Applies Dropout to the input.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.BatchNorm" title="mxnet.gluon.nn.BatchNorm"><code class="xref py py-obj docutils literal">BatchNorm</code></a></td> <td>Batch normalization layer (Ioffe and Szegedy, 2014).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.InstanceNorm" title="mxnet.gluon.nn.InstanceNorm"><code class="xref py py-obj docutils literal">InstanceNorm</code></a></td> <td>Applies instance normalization to the n-dimensional input array.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.LayerNorm" title="mxnet.gluon.nn.LayerNorm"><code class="xref py py-obj docutils literal">LayerNorm</code></a></td> <td>Applies layer normalization to the n-dimensional input array.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Embedding" title="mxnet.gluon.nn.Embedding"><code class="xref py py-obj docutils literal">Embedding</code></a></td> <td>Turns non-negative integers (indexes/tokens) into dense vectors of fixed size.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.Flatten" title="mxnet.gluon.nn.Flatten"><code class="xref py py-obj docutils literal">Flatten</code></a></td> <td>Flattens the input to two dimensional.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Lambda" title="mxnet.gluon.nn.Lambda"><code class="xref py py-obj docutils literal">Lambda</code></a></td> <td>Wraps an operator or an expression as a Block object.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.HybridLambda" title="mxnet.gluon.nn.HybridLambda"><code class="xref py py-obj docutils literal">HybridLambda</code></a></td> <td>Wraps an operator or an expression as a HybridBlock object.</td> </tr> </tbody> </table> </div> <div class="section" id="convolutional-layers"> <h2>Convolutional Layers<a class="headerlink" href="#convolutional-layers" 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.nn.Conv1D" title="mxnet.gluon.nn.Conv1D"><code class="xref py py-obj docutils literal">Conv1D</code></a></td> <td>1D convolution layer (e.g. temporal convolution).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Conv2D" title="mxnet.gluon.nn.Conv2D"><code class="xref py py-obj docutils literal">Conv2D</code></a></td> <td>2D convolution layer (e.g. spatial convolution over images).</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.Conv3D" title="mxnet.gluon.nn.Conv3D"><code class="xref py py-obj docutils literal">Conv3D</code></a></td> <td>3D convolution layer (e.g. spatial convolution over volumes).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Conv1DTranspose" title="mxnet.gluon.nn.Conv1DTranspose"><code class="xref py py-obj docutils literal">Conv1DTranspose</code></a></td> <td>Transposed 1D convolution layer (sometimes called Deconvolution).</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.Conv2DTranspose" title="mxnet.gluon.nn.Conv2DTranspose"><code class="xref py py-obj docutils literal">Conv2DTranspose</code></a></td> <td>Transposed 2D convolution layer (sometimes called Deconvolution).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Conv3DTranspose" title="mxnet.gluon.nn.Conv3DTranspose"><code class="xref py py-obj docutils literal">Conv3DTranspose</code></a></td> <td>Transposed 3D convolution layer (sometimes called Deconvolution).</td> </tr> </tbody> </table> </div> <div class="section" id="pooling-layers"> <h2>Pooling Layers<a class="headerlink" href="#pooling-layers" 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.nn.MaxPool1D" title="mxnet.gluon.nn.MaxPool1D"><code class="xref py py-obj docutils literal">MaxPool1D</code></a></td> <td>Max pooling operation for one dimensional data.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.MaxPool2D" title="mxnet.gluon.nn.MaxPool2D"><code class="xref py py-obj docutils literal">MaxPool2D</code></a></td> <td>Max pooling operation for two dimensional (spatial) data.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.MaxPool3D" title="mxnet.gluon.nn.MaxPool3D"><code class="xref py py-obj docutils literal">MaxPool3D</code></a></td> <td>Max pooling operation for 3D data (spatial or spatio-temporal).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.AvgPool1D" title="mxnet.gluon.nn.AvgPool1D"><code class="xref py py-obj docutils literal">AvgPool1D</code></a></td> <td>Average pooling operation for temporal data.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.AvgPool2D" title="mxnet.gluon.nn.AvgPool2D"><code class="xref py py-obj docutils literal">AvgPool2D</code></a></td> <td>Average pooling operation for spatial data.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.AvgPool3D" title="mxnet.gluon.nn.AvgPool3D"><code class="xref py py-obj docutils literal">AvgPool3D</code></a></td> <td>Average pooling operation for 3D data (spatial or spatio-temporal).</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalMaxPool1D" title="mxnet.gluon.nn.GlobalMaxPool1D"><code class="xref py py-obj docutils literal">GlobalMaxPool1D</code></a></td> <td>Gloabl max pooling operation for one dimensional (temporal) data.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalMaxPool2D" title="mxnet.gluon.nn.GlobalMaxPool2D"><code class="xref py py-obj docutils literal">GlobalMaxPool2D</code></a></td> <td>Global max pooling operation for two dimensional (spatial) data.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalMaxPool3D" title="mxnet.gluon.nn.GlobalMaxPool3D"><code class="xref py py-obj docutils literal">GlobalMaxPool3D</code></a></td> <td>Global max pooling operation for 3D data (spatial or spatio-temporal).</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalAvgPool1D" title="mxnet.gluon.nn.GlobalAvgPool1D"><code class="xref py py-obj docutils literal">GlobalAvgPool1D</code></a></td> <td>Global average pooling operation for temporal data.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalAvgPool2D" title="mxnet.gluon.nn.GlobalAvgPool2D"><code class="xref py py-obj docutils literal">GlobalAvgPool2D</code></a></td> <td>Global average pooling operation for spatial data.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.GlobalAvgPool3D" title="mxnet.gluon.nn.GlobalAvgPool3D"><code class="xref py py-obj docutils literal">GlobalAvgPool3D</code></a></td> <td>Global average pooling operation for 3D data (spatial or spatio-temporal).</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.ReflectionPad2D" title="mxnet.gluon.nn.ReflectionPad2D"><code class="xref py py-obj docutils literal">ReflectionPad2D</code></a></td> <td>Pads the input tensor using the reflection of the input boundary.</td> </tr> </tbody> </table> </div> <div class="section" id="activation-layers"> <h2>Activation Layers<a class="headerlink" href="#activation-layers" 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.nn.Activation" title="mxnet.gluon.nn.Activation"><code class="xref py py-obj docutils literal">Activation</code></a></td> <td>Applies an activation function to input.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.LeakyReLU" title="mxnet.gluon.nn.LeakyReLU"><code class="xref py py-obj docutils literal">LeakyReLU</code></a></td> <td>Leaky version of a Rectified Linear Unit.</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.PReLU" title="mxnet.gluon.nn.PReLU"><code class="xref py py-obj docutils literal">PReLU</code></a></td> <td>Parametric leaky version of a Rectified Linear Unit.</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.ELU" title="mxnet.gluon.nn.ELU"><code class="xref py py-obj docutils literal">ELU</code></a></td> <td>Exponential Linear Unit (ELU)</td> </tr> <tr class="row-odd"><td><a class="reference internal" href="#mxnet.gluon.nn.SELU" title="mxnet.gluon.nn.SELU"><code class="xref py py-obj docutils literal">SELU</code></a></td> <td>Scaled Exponential Linear Unit (SELU)</td> </tr> <tr class="row-even"><td><a class="reference internal" href="#mxnet.gluon.nn.Swish" title="mxnet.gluon.nn.Swish"><code class="xref py py-obj docutils literal">Swish</code></a></td> <td>Swish Activation function</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 layers. <dl class="class"> <dt id="mxnet.gluon.nn.Activation"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Activation</code>(activation, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#Activation">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Activation" title="Permalink to this definition">¶</a></dt> <dd>Applies an activation function to input. <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">activation (str) – Name of activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a> for available choices.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.AvgPool1D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">AvgPool1D</code>(pool_size=2, strides=None, padding=0, layout='NCW', ceil_mode=False, count_include_pad=True, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#AvgPool1D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.AvgPool1D" title="Permalink to this definition">¶</a></dt> <dd>Average pooling operation for temporal data. <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>pool_size (int) – Size of the max pooling windows.</li> <li>strides (int, or None) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. padding is applied on ‘W’ dimension.</li> <li>ceil_mode (bool, default False) – When <cite>True</cite>, will use ceil instead of floor to compute the output shape.</li> <li>count_include_pad (bool, default True) – When ‘False’, will exclude padding elements when computing the average value.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, out_width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. out_width is calculated as: <div class="highlight-default"><div class="highlight"><pre>out_width = floor((width+2*padding-pool_size)/strides)+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True</cite>, ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.AvgPool2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">AvgPool2D</code>(pool_size=(2, 2), strides=None, padding=0, ceil_mode=False, layout='NCHW', count_include_pad=True, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#AvgPool2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.AvgPool2D" title="Permalink to this definition">¶</a></dt> <dd>Average pooling operation for spatial data. <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>pool_size (int or list/tuple of 2 ints,) – Size of the max pooling windows.</li> <li>strides (int, list/tuple of 2 ints, or None.) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int or list/tuple of 2 ints,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively. padding is applied on ‘H’ and ‘W’ dimension.</li> <li>ceil_mode (bool, default False) – When True, will use ceil instead of floor to compute the output shape.</li> <li>count_include_pad (bool, default True) – When ‘False’, will exclude padding elements when computing the average value.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_height = floor((height+2*padding[0]-pool_size[0])/strides[0])+1 out_width = floor((width+2*padding[1]-pool_size[1])/strides[1])+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True</cite>, ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.AvgPool3D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">AvgPool3D</code>(pool_size=(2, 2, 2), strides=None, padding=0, ceil_mode=False, layout='NCDHW', count_include_pad=True, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#AvgPool3D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.AvgPool3D" title="Permalink to this definition">¶</a></dt> <dd>Average pooling operation for 3D data (spatial or spatio-temporal). <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>pool_size (int or list/tuple of 3 ints,) – Size of the max pooling windows.</li> <li>strides (int, list/tuple of 3 ints, or None.) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int or list/tuple of 3 ints,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCDHW') – Dimension ordering of data and weight. Can be ‘NCDHW’, ‘NDHWC’, etc. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. padding is applied on ‘D’, ‘H’ and ‘W’ dimension.</li> <li>ceil_mode (bool, default False) – When True, will use ceil instead of floor to compute the output shape.</li> <li>count_include_pad (bool, default True) – When ‘False’, will exclude padding elements when computing the average value.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, out_depth, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. out_depth, out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_depth = floor((depth+2*padding[0]-pool_size[0])/strides[0])+1 out_height = floor((height+2*padding[1]-pool_size[1])/strides[1])+1 out_width = floor((width+2*padding[2]-pool_size[2])/strides[2])+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True,</cite> ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.BatchNorm"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">BatchNorm</code>(axis=1, momentum=0.9, epsilon=1e-05, center=True, scale=True, use_global_stats=False, beta_initializer='zeros', gamma_initializer='ones', running_mean_initializer='zeros', running_variance_initializer='ones', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#BatchNorm">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.BatchNorm" title="Permalink to this definition">¶</a></dt> <dd>Batch normalization layer (Ioffe and Szegedy, 2014). Normalizes the input at each batch, i.e. applies a transformation that maintains the mean activation close to 0 and the activation standard deviation close to 1. <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>axis (int, default 1) – The axis that should be normalized. This is typically the channels (C) axis. For instance, after a <cite>Conv2D</cite> layer with <cite>layout=’NCHW’</cite>, set <cite>axis=1</cite> in <cite>BatchNorm</cite>. If <cite>layout=’NHWC’</cite>, then set <cite>axis=3</cite>.</li> <li>momentum (float, default 0.9) – Momentum for the moving average.</li> <li>epsilon (float, default 1e-5) – Small float added to variance to avoid dividing by zero.</li> <li>center (bool, default True) – If True, add offset of <cite>beta</cite> to normalized tensor. If False, <cite>beta</cite> is ignored.</li> <li>scale (bool, default True) – If True, multiply by <cite>gamma</cite>. If False, <cite>gamma</cite> is not used. When the next layer is linear (also e.g. <cite>nn.relu</cite>), this can be disabled since the scaling will be done by the next layer.</li> <li>use_global_stats (bool, default False) – If True, use global moving statistics instead of local batch-norm. This will force change batch-norm into a scale shift operator. If False, use local batch-norm.</li> <li>beta_initializer (str or <cite>Initializer</cite>, default ‘zeros’) – Initializer for the beta weight.</li> <li>gamma_initializer (str or <cite>Initializer</cite>, default ‘ones’) – Initializer for the gamma weight.</li> <li>moving_mean_initializer (str or <cite>Initializer</cite>, default ‘zeros’) – Initializer for the moving mean.</li> <li>moving_variance_initializer (str or <cite>Initializer</cite>, default ‘ones’) – Initializer for the moving variance.</li> <li>in_channels (int, default 0) – Number of channels (feature maps) in input data. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv1D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv1D</code>(channels, kernel_size, strides=1, padding=0, dilation=1, groups=1, layout='NCW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv1D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv1D" title="Permalink to this definition">¶</a></dt> <dd>1D convolution layer (e.g. temporal convolution). This layer creates a convolution kernel that is convolved with the layer input over a single spatial (or temporal) dimension to produce a tensor of outputs. If <cite>use_bias</cite> is True, a bias vector is created and added to the outputs. Finally, if <cite>activation</cite> is not <cite>None</cite>, it is applied to the outputs as well. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 1 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 1 int,) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 1 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>dilation (int or tuple/list of 1 int) – Specifies the dilation rate to use for dilated convolution.</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. Convolution is applied on the ‘W’ dimension.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, out_width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. out_width is calculated as: <div class="highlight-default"><div class="highlight"><pre>out_width = floor((width+2*padding-dilation*(kernel_size-1)-1)/stride)+1 </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv1DTranspose"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv1DTranspose</code>(channels, kernel_size, strides=1, padding=0, output_padding=0, dilation=1, groups=1, layout='NCW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv1DTranspose">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv1DTranspose" title="Permalink to this definition">¶</a></dt> <dd>Transposed 1D convolution layer (sometimes called Deconvolution). The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 1 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 1 int) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 1 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>output_padding (int or a tuple/list of 1 int) – Controls the amount of implicit zero-paddings on both sides of the output for output_padding number of points for each dimension.</li> <li>dilation (int or tuple/list of 1 int) – Controls the spacing between the kernel points; also known as the a trous algorithm</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. Convolution is applied on the ‘W’ dimension.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, out_width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. out_width is calculated as: <div class="highlight-default"><div class="highlight"><pre>out_width = (width-1)*strides-2*padding+kernel_size+output_padding </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv2D</code>(channels, kernel_size, strides=(1, 1), padding=(0, 0), dilation=(1, 1), groups=1, layout='NCHW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv2D" title="Permalink to this definition">¶</a></dt> <dd>2D convolution layer (e.g. spatial convolution over images). This layer creates a convolution kernel that is convolved with the layer input to produce a tensor of outputs. If <cite>use_bias</cite> is True, a bias vector is created and added to the outputs. Finally, if <cite>activation</cite> is not <cite>None</cite>, it is applied to the outputs as well. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 2 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 2 int,) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 2 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>dilation (int or tuple/list of 2 int) – Specifies the dilation rate to use for dilated convolution.</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ and ‘NHWC’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively. Convolution is applied on the ‘H’ and ‘W’ dimensions.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_height = floor((height+2*padding[0]-dilation[0]*(kernel_size[0]-1)-1)/stride[0])+1 out_width = floor((width+2*padding[1]-dilation[1]*(kernel_size[1]-1)-1)/stride[1])+1 </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv2DTranspose"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv2DTranspose</code>(channels, kernel_size, strides=(1, 1), padding=(0, 0), output_padding=(0, 0), dilation=(1, 1), groups=1, layout='NCHW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv2DTranspose">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv2DTranspose" title="Permalink to this definition">¶</a></dt> <dd>Transposed 2D convolution layer (sometimes called Deconvolution). The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 2 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 2 int) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 2 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>output_padding (int or a tuple/list of 2 int) – Controls the amount of implicit zero-paddings on both sides of the output for output_padding number of points for each dimension.</li> <li>dilation (int or tuple/list of 2 int) – Controls the spacing between the kernel points; also known as the a trous algorithm</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ and ‘NHWC’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively. Convolution is applied on the ‘H’ and ‘W’ dimensions.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_height = (height-1)*strides[0]-2*padding[0]+kernel_size[0]+output_padding[0] out_width = (width-1)*strides[1]-2*padding[1]+kernel_size[1]+output_padding[1] </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv3D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv3D</code>(channels, kernel_size, strides=(1, 1, 1), padding=(0, 0, 0), dilation=(1, 1, 1), groups=1, layout='NCDHW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv3D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv3D" title="Permalink to this definition">¶</a></dt> <dd>3D convolution layer (e.g. spatial convolution over volumes). This layer creates a convolution kernel that is convolved with the layer input to produce a tensor of outputs. If <cite>use_bias</cite> is <cite>True</cite>, a bias vector is created and added to the outputs. Finally, if <cite>activation</cite> is not <cite>None</cite>, it is applied to the outputs as well. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 3 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 3 int,) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 3 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>dilation (int or tuple/list of 3 int) – Specifies the dilation rate to use for dilated convolution.</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCDHW') – Dimension ordering of data and weight. Only supports ‘NCDHW’ and ‘NDHWC’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. Convolution is applied on the ‘D’, ‘H’ and ‘W’ dimensions.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, out_depth, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. out_depth, out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_depth = floor((depth+2*padding[0]-dilation[0]*(kernel_size[0]-1)-1)/stride[0])+1 out_height = floor((height+2*padding[1]-dilation[1]*(kernel_size[1]-1)-1)/stride[1])+1 out_width = floor((width+2*padding[2]-dilation[2]*(kernel_size[2]-1)-1)/stride[2])+1 </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Conv3DTranspose"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Conv3DTranspose</code>(channels, kernel_size, strides=(1, 1, 1), padding=(0, 0, 0), output_padding=(0, 0, 0), dilation=(1, 1, 1), groups=1, layout='NCDHW', activation=None, use_bias=True, weight_initializer=None, bias_initializer='zeros', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#Conv3DTranspose">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Conv3DTranspose" title="Permalink to this definition">¶</a></dt> <dd>Transposed 3D convolution layer (sometimes called Deconvolution). The need for transposed convolutions generally arises from the desire to use a transformation going in the opposite direction of a normal convolution, i.e., from something that has the shape of the output of some convolution to something that has the shape of its input while maintaining a connectivity pattern that is compatible with said convolution. If <cite>in_channels</cite> is not specified, <cite>Parameter</cite> initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data. <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>channels (int) – The dimensionality of the output space, i.e. the number of output channels (filters) in the convolution.</li> <li>kernel_size (int or tuple/list of 3 int) – Specifies the dimensions of the convolution window.</li> <li>strides (int or tuple/list of 3 int) – Specify the strides of the convolution.</li> <li>padding (int or a tuple/list of 3 int,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points</li> <li>output_padding (int or a tuple/list of 3 int) – Controls the amount of implicit zero-paddings on both sides of the output for output_padding number of points for each dimension.</li> <li>dilation (int or tuple/list of 3 int) – Controls the spacing between the kernel points; also known as the a trous algorithm.</li> <li>groups (int) – Controls the connections between inputs and outputs. At groups=1, all inputs are convolved to all outputs. At groups=2, the operation becomes equivalent to having two conv layers side by side, each seeing half the input channels, and producing half the output channels, and both subsequently concatenated.</li> <li>layout (str, default 'NCDHW') – Dimension ordering of data and weight. Only supports ‘NCDHW’ and ‘NDHWC’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. Convolution is applied on the ‘D’, ‘H’ and ‘W’ dimensions.</li> <li>in_channels (int, default 0) – The number of input channels to this layer. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> <li>activation (str) – Activation function to use. See <a class="reference internal" href="../ndarray/ndarray.html#mxnet.ndarray.Activation" title="mxnet.ndarray.Activation"><code class="xref py py-func docutils literal">Activation()</code></a>. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool) – Whether the layer uses a bias vector.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>weight</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, out_depth, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. out_depth, out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_depth = (depth-1)*strides[0]-2*padding[0]+kernel_size[0]+output_padding[0] out_height = (height-1)*strides[1]-2*padding[1]+kernel_size[1]+output_padding[1] out_width = (width-1)*strides[2]-2*padding[2]+kernel_size[2]+output_padding[2] </pre></div> </div> </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Dense"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Dense</code>(units, activation=None, use_bias=True, flatten=True, dtype='float32', weight_initializer=None, bias_initializer='zeros', in_units=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Dense">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Dense" title="Permalink to this definition">¶</a></dt> <dd>Just your regular densely-connected NN layer. <cite>Dense</cite> implements the operation: <cite>output = activation(dot(input, weight) + bias)</cite> where <cite>activation</cite> is the element-wise activation function passed as the <cite>activation</cite> argument, <cite>weight</cite> is a weights matrix created by the layer, and <cite>bias</cite> is a bias vector created by the layer (only applicable if <cite>use_bias</cite> is <cite>True</cite>). Note: the input must be a tensor with rank 2. Use <cite>flatten</cite> to convert it to rank 2 manually if necessary. <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>units (int) – Dimensionality of the output space.</li> <li>activation (str) – Activation function to use. See help on <cite>Activation</cite> layer. If you don’t specify anything, no activation is applied (ie. “linear” activation: <cite>a(x) = x</cite>).</li> <li>use_bias (bool, default True) – Whether the layer uses a bias vector.</li> <li>flatten (bool, default True) – Whether the input tensor should be flattened. If true, all but the first axis of input data are collapsed together. If false, all but the last axis of input data are kept the same, and the transformation applies on the last axis.</li> <li>dtype (str or np.dtype, default 'float32') – Data type of output embeddings.</li> <li>weight_initializer (str or <cite>Initializer</cite>) – Initializer for the <cite>kernel</cite> weights matrix.</li> <li>bias_initializer (str or <cite>Initializer</cite>) – Initializer for the bias vector.</li> <li>in_units (int, optional) – Size of the input data. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_units</cite> will be inferred from the shape of input data.</li> <li>prefix (str or None) – See document of <cite>Block</cite>.</li> <li>params (<a class="reference internal" href="gluon.html#mxnet.gluon.ParameterDict" title="mxnet.gluon.ParameterDict">ParameterDict</a> or None) – See document of <cite>Block</cite>.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: if <cite>flatten</cite> is True, <cite>data</cite> should be a tensor with shape <cite>(batch_size, x1, x2, ..., xn)</cite>, where x1 * x2 * ... * xn is equal to <cite>in_units</cite>. If <cite>flatten</cite> is False, <cite>data</cite> should have shape <cite>(x1, x2, ..., xn, in_units)</cite>.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: if <cite>flatten</cite> is True, <cite>out</cite> will be a tensor with shape <cite>(batch_size, units)</cite>. If <cite>flatten</cite> is False, <cite>out</cite> will have shape <cite>(x1, x2, ..., xn, units)</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Dropout"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Dropout</code>(rate, axes=(), **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Dropout">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Dropout" title="Permalink to this definition">¶</a></dt> <dd>Applies Dropout to the input. Dropout consists in randomly setting a fraction <cite>rate</cite> of input units to 0 at each update during training time, which helps prevent overfitting. <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>rate (float) – Fraction of the input units to drop. Must be a number between 0 and 1.</li> <li>axes (tuple of int, default ()) – The axes on which dropout mask is shared. If empty, regular dropout is applied.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> References <a class="reference external" href="http://www.cs.toronto.edu/~rsalakhu/papers/srivastava14a.pdf">Dropout: A Simple Way to Prevent Neural Networks from Overfitting</a> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.ELU"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">ELU</code>(alpha=1.0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#ELU">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.ELU" title="Permalink to this definition">¶</a></dt> <dd><dl class="docutils"> <dt>Exponential Linear Unit (ELU)</dt> <dd>“Fast and Accurate Deep Network Learning by Exponential Linear Units”, Clevert et al, 2016 <a class="reference external" href="https://arxiv.org/abs/1511.07289">https://arxiv.org/abs/1511.07289</a> Published as a conference paper at ICLR 2016</dd> </dl> <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">alpha (float) – The alpha parameter as described by Clevert et al, 2016</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Embedding"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Embedding</code>(input_dim, output_dim, dtype='float32', weight_initializer=None, sparse_grad=False, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Embedding">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Embedding" title="Permalink to this definition">¶</a></dt> <dd>Turns non-negative integers (indexes/tokens) into dense vectors of fixed size. eg. [4, 20] -> [[0.25, 0.1], [0.6, -0.2]] Note: if <cite>sparse_grad</cite> is set to True, the gradient w.r.t weight will be sparse. Only a subset of optimizers support sparse gradients, including SGD, AdaGrad and Adam. By default lazy updates is turned on, which may perform differently from standard updates. For more details, please check the Optimization API at: <a class="reference external" href="/api/python/optimization/optimization.html">/api/python/optimization/optimization.html</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>input_dim (int) – Size of the vocabulary, i.e. maximum integer index + 1.</li> <li>output_dim (int) – Dimension of the dense embedding.</li> <li>dtype (str or np.dtype, default 'float32') – Data type of output embeddings.</li> <li>weight_initializer (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – Initializer for the <cite>embeddings</cite> matrix.</li> <li>sparse_grad (bool) – If True, gradient w.r.t. weight will be a ‘row_sparse’ NDArray.</li> <li>Inputs – <ul> <li>data: (N-1)-D tensor with shape: <cite>(x1, x2, ..., xN-1)</cite>.</li> </ul> </li> <li>Output – <ul> <li>out: N-D tensor with shape: <cite>(x1, x2, ..., xN-1, output_dim)</cite>.</li> </ul> </li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Flatten"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Flatten</code>(**kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Flatten">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Flatten" title="Permalink to this definition">¶</a></dt> <dd>Flattens the input to two dimensional. <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape <cite>(N, x1, x2, ..., xn)</cite></li> </ul> </dd> <dt>Output:</dt> <dd><ul class="first last simple"> <li>out: 2D tensor with shape: <cite>(N, x1 cdot x2 cdot ... cdot xn)</cite></li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalAvgPool1D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalAvgPool1D</code>(layout='NCW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalAvgPool1D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalAvgPool1D" title="Permalink to this definition">¶</a></dt> <dd>Global average pooling operation for temporal data. <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">layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. padding is applied on ‘W’ dimension.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, 1)</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalAvgPool2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalAvgPool2D</code>(layout='NCHW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalAvgPool2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalAvgPool2D" title="Permalink to this definition">¶</a></dt> <dd>Global average pooling operation for spatial data. <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">layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, 1, 1)</cite> when <cite>layout</cite> is <cite>NCHW</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalAvgPool3D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalAvgPool3D</code>(layout='NCDHW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalAvgPool3D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalAvgPool3D" title="Permalink to this definition">¶</a></dt> <dd>Global average pooling operation for 3D data (spatial or spatio-temporal). <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">layout (str, default 'NCDHW') – Dimension ordering of data and weight. Can be ‘NCDHW’, ‘NDHWC’, etc. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. padding is applied on ‘D’, ‘H’ and ‘W’ dimension.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, 1, 1, 1)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalMaxPool1D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalMaxPool1D</code>(layout='NCW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalMaxPool1D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalMaxPool1D" title="Permalink to this definition">¶</a></dt> <dd>Gloabl max pooling operation for one dimensional (temporal) data. <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">layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. Pooling is applied on the W dimension.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, 1)</cite> when <cite>layout</cite> is <cite>NCW</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalMaxPool2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalMaxPool2D</code>(layout='NCHW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalMaxPool2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalMaxPool2D" title="Permalink to this definition">¶</a></dt> <dd>Global max pooling operation for two dimensional (spatial) data. <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">layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively. padding is applied on ‘H’ and ‘W’ dimension.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, 1, 1)</cite> when <cite>layout</cite> is <cite>NCHW</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.GlobalMaxPool3D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">GlobalMaxPool3D</code>(layout='NCDHW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#GlobalMaxPool3D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.GlobalMaxPool3D" title="Permalink to this definition">¶</a></dt> <dd>Global max pooling operation for 3D data (spatial or spatio-temporal). <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">layout (str, default 'NCDHW') – Dimension ordering of data and weight. Only supports ‘NCDHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. padding is applied on ‘D’, ‘H’ and ‘W’ dimension.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, 1, 1, 1)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.HybridLambda"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">HybridLambda</code>(function, prefix=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#HybridLambda">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.HybridLambda" title="Permalink to this definition">¶</a></dt> <dd>Wraps an operator or an expression as a HybridBlock object. <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>function (str or function) – Function used in lambda must be one of the following: 1) The name of an operator that is available in both symbol and ndarray. For example: <div class="highlight-default"><div class="highlight"><pre>block = HybridLambda('tanh') </pre></div> </div> <ol class="arabic" start="2"> <li>A function that conforms to <code class="docutils literal">def function(F, data, *args)</code>. For example:<div class="highlight-default"><div class="highlight"><pre>block = HybridLambda(lambda F, x: F.LeakyReLU(x, slope=0.1)) </pre></div> </div> </li> </ol> </li> <li>Inputs – <ul> <li><dl class="first docutils"> <dt>** args *: one or more input data. First argument must be symbol or ndarray. Their </dt> <dd>shapes depend on the function.</dd> </dl> </li> </ul> </li> <li>Output – <ul> <li>** outputs *: one or more output data. Their shapes depend on the function.</li> </ul> </li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.InstanceNorm"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">InstanceNorm</code>(axis=1, epsilon=1e-05, center=True, scale=False, beta_initializer='zeros', gamma_initializer='ones', in_channels=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#InstanceNorm">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.InstanceNorm" title="Permalink to this definition">¶</a></dt> <dd>Applies instance normalization to the n-dimensional input array. This operator takes an n-dimensional input array where (n>2) and normalizes the input using the following formula: <div class="math"> \[ \begin{align}\begin{aligned}\bar{C} = \{i \mid i \neq 0, i \neq axis\}\\out = \frac{x - mean[data, \bar{C}]}{ \sqrt{Var[data, \bar{C}]} + \epsilon} * gamma + beta\end{aligned}\end{align} \]</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>axis (int, default 1) – The axis that will be excluded in the normalization process. This is typically the channels (C) axis. For instance, after a <cite>Conv2D</cite> layer with <cite>layout=’NCHW’</cite>, set <cite>axis=1</cite> in <cite>InstanceNorm</cite>. If <cite>layout=’NHWC’</cite>, then set <cite>axis=3</cite>. Data will be normalized along axes excluding the first axis and the axis given.</li> <li>epsilon (float, default 1e-5) – Small float added to variance to avoid dividing by zero.</li> <li>center (bool, default True) – If True, add offset of <cite>beta</cite> to normalized tensor. If False, <cite>beta</cite> is ignored.</li> <li>scale (bool, default True) – If True, multiply by <cite>gamma</cite>. If False, <cite>gamma</cite> is not used. When the next layer is linear (also e.g. <cite>nn.relu</cite>), this can be disabled since the scaling will be done by the next layer.</li> <li>beta_initializer (str or <cite>Initializer</cite>, default ‘zeros’) – Initializer for the beta weight.</li> <li>gamma_initializer (str or <cite>Initializer</cite>, default ‘ones’) – Initializer for the gamma weight.</li> <li>in_channels (int, default 0) – Number of channels (feature maps) in input data. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> References <a class="reference external" href="https://arxiv.org/abs/1607.08022">Instance Normalization: The Missing Ingredient for Fast Stylization</a> Examples <div class="highlight-default"><div class="highlight"><pre>>>> # Input of shape (2,1,2) >>> x = mx.nd.array([[[ 1.1, 2.2]], ... [[ 3.3, 4.4]]]) >>> # Instance normalization is calculated with the above formula >>> layer = InstanceNorm() >>> layer.initialize(ctx=mx.cpu(0)) >>> layer(x) [[[-0.99998355 0.99998331]] [[-0.99998319 0.99998361]]] <NDArray 2x1x2 @cpu(0)> </pre></div> </div> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Lambda"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Lambda</code>(function, prefix=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#Lambda">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Lambda" title="Permalink to this definition">¶</a></dt> <dd>Wraps an operator or an expression as a Block object. <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>function (str or function) – Function used in lambda must be one of the following: 1) the name of an operator that is available in ndarray. For example: <div class="highlight-default"><div class="highlight"><pre>block = Lambda('tanh') </pre></div> </div> <ol class="arabic" start="2"> <li>a function that conforms to <code class="docutils literal">def function(*args)</code>. For example:<div class="highlight-default"><div class="highlight"><pre>block = Lambda(lambda x: nd.LeakyReLU(x, slope=0.1)) </pre></div> </div> </li> </ol> </li> <li>Inputs – <ul> <li>** args *: one or more input data. Their shapes depend on the function.</li> </ul> </li> <li>Output – <ul> <li>** outputs *: one or more output data. Their shapes depend on the function.</li> </ul> </li> </ul> </td> </tr> </tbody> </table> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.LayerNorm"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">LayerNorm</code>(axis=-1, epsilon=1e-05, center=True, scale=True, beta_initializer='zeros', gamma_initializer='ones', in_channels=0, prefix=None, params=None)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/basic_layers.html#LayerNorm">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.LayerNorm" title="Permalink to this definition">¶</a></dt> <dd>Applies layer normalization to the n-dimensional input array. This operator takes an n-dimensional input array and normalizes the input using the given axis: <div class="math"> \[out = \frac{x - mean[data, axis]}{ \sqrt{Var[data, axis]} + \epsilon} * gamma + beta\]</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>axis (int, default -1) – The axis that should be normalized. This is typically the axis of the channels.</li> <li>epsilon (float, default 1e-5) – Small float added to variance to avoid dividing by zero.</li> <li>center (bool, default True) – If True, add offset of <cite>beta</cite> to normalized tensor. If False, <cite>beta</cite> is ignored.</li> <li>scale (bool, default True) – If True, multiply by <cite>gamma</cite>. If False, <cite>gamma</cite> is not used.</li> <li>beta_initializer (str or <cite>Initializer</cite>, default ‘zeros’) – Initializer for the beta weight.</li> <li>gamma_initializer (str or <cite>Initializer</cite>, default ‘ones’) – Initializer for the gamma weight.</li> <li>in_channels (int, default 0) – Number of channels (feature maps) in input data. If not specified, initialization will be deferred to the first time <cite>forward</cite> is called and <cite>in_channels</cite> will be inferred from the shape of input data.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> References <a class="reference external" href="https://arxiv.org/pdf/1607.06450.pdf">Layer Normalization</a> Examples <div class="highlight-default"><div class="highlight"><pre>>>> # Input of shape (2, 5) >>> x = mx.nd.array([[1, 2, 3, 4, 5], [1, 1, 2, 2, 2]]) >>> # Layer normalization is calculated with the above formula >>> layer = LayerNorm() >>> layer.initialize(ctx=mx.cpu(0)) >>> layer(x) [[-1.41421 -0.707105 0. 0.707105 1.41421 ] [-1.2247195 -1.2247195 0.81647956 0.81647956 0.81647956]] <NDArray 2x5 @cpu(0)> </pre></div> </div> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.LeakyReLU"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">LeakyReLU</code>(alpha, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#LeakyReLU">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.LeakyReLU" title="Permalink to this definition">¶</a></dt> <dd>Leaky version of a Rectified Linear Unit. It allows a small gradient when the unit is not active <div class="math"> \[\begin{split}f\left(x\right) = \left\{ \begin{array}{lr} \alpha x & : x \lt 0 \\ x & : x \geq 0 \\ \end{array} \right.\\\end{split}\]</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">alpha (float) – slope coefficient for the negative half axis. Must be >= 0.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.MaxPool1D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">MaxPool1D</code>(pool_size=2, strides=None, padding=0, layout='NCW', ceil_mode=False, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#MaxPool1D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.MaxPool1D" title="Permalink to this definition">¶</a></dt> <dd>Max pooling operation for one dimensional data. <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>pool_size (int) – Size of the max pooling windows.</li> <li>strides (int, or None) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCW') – Dimension ordering of data and weight. Only supports ‘NCW’ layout for now. ‘N’, ‘C’, ‘W’ stands for batch, channel, and width (time) dimensions respectively. Pooling is applied on the W dimension.</li> <li>ceil_mode (bool, default False) – When <cite>True</cite>, will use ceil instead of floor to compute the output shape.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 3D input tensor with shape <cite>(batch_size, in_channels, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 3D output tensor with shape <cite>(batch_size, channels, out_width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. out_width is calculated as: <div class="highlight-default"><div class="highlight"><pre>out_width = floor((width+2*padding-pool_size)/strides)+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True</cite>, ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.MaxPool2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">MaxPool2D</code>(pool_size=(2, 2), strides=None, padding=0, layout='NCHW', ceil_mode=False, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#MaxPool2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.MaxPool2D" title="Permalink to this definition">¶</a></dt> <dd>Max pooling operation for two dimensional (spatial) data. <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>pool_size (int or list/tuple of 2 ints,) – Size of the max pooling windows.</li> <li>strides (int, list/tuple of 2 ints, or None.) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int or list/tuple of 2 ints,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCHW') – Dimension ordering of data and weight. Only supports ‘NCHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’ stands for batch, channel, height, and width dimensions respectively. padding is applied on ‘H’ and ‘W’ dimension.</li> <li>ceil_mode (bool, default False) – When <cite>True</cite>, will use ceil instead of floor to compute the output shape.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 4D input tensor with shape <cite>(batch_size, in_channels, height, width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 4D output tensor with shape <cite>(batch_size, channels, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCHW</cite>. out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_height = floor((height+2*padding[0]-pool_size[0])/strides[0])+1 out_width = floor((width+2*padding[1]-pool_size[1])/strides[1])+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True</cite>, ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.MaxPool3D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">MaxPool3D</code>(pool_size=(2, 2, 2), strides=None, padding=0, ceil_mode=False, layout='NCDHW', **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#MaxPool3D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.MaxPool3D" title="Permalink to this definition">¶</a></dt> <dd>Max pooling operation for 3D data (spatial or spatio-temporal). <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>pool_size (int or list/tuple of 3 ints,) – Size of the max pooling windows.</li> <li>strides (int, list/tuple of 3 ints, or None.) – Factor by which to downscale. E.g. 2 will halve the input size. If <cite>None</cite>, it will default to <cite>pool_size</cite>.</li> <li>padding (int or list/tuple of 3 ints,) – If padding is non-zero, then the input is implicitly zero-padded on both sides for padding number of points.</li> <li>layout (str, default 'NCDHW') – Dimension ordering of data and weight. Only supports ‘NCDHW’ layout for now. ‘N’, ‘C’, ‘H’, ‘W’, ‘D’ stands for batch, channel, height, width and depth dimensions respectively. padding is applied on ‘D’, ‘H’ and ‘W’ dimension.</li> <li>ceil_mode (bool, default False) – When <cite>True</cite>, will use ceil instead of floor to compute the output shape.</li> </ul> </td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: 5D input tensor with shape <cite>(batch_size, in_channels, depth, height, width)</cite> when <cite>layout</cite> is <cite>NCW</cite>. For other layouts shape is permuted accordingly.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: 5D output tensor with shape <cite>(batch_size, channels, out_depth, out_height, out_width)</cite> when <cite>layout</cite> is <cite>NCDHW</cite>. out_depth, out_height and out_width are calculated as: <div class="highlight-default"><div class="highlight"><pre>out_depth = floor((depth+2*padding[0]-pool_size[0])/strides[0])+1 out_height = floor((height+2*padding[1]-pool_size[1])/strides[1])+1 out_width = floor((width+2*padding[2]-pool_size[2])/strides[2])+1 </pre></div> </div> When <cite>ceil_mode</cite> is <cite>True</cite>, ceil will be used instead of floor in this equation. </li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.PReLU"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">PReLU</code>(alpha_initializer=<mxnet.initializer.Constant object>, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#PReLU">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.PReLU" title="Permalink to this definition">¶</a></dt> <dd>Parametric leaky version of a Rectified Linear Unit. <<a class="reference external" href="https://arxiv.org/abs/1502.01852">https://arxiv.org/abs/1502.01852</a>>`_ paper. It learns a gradient when the unit is not active <div class="math"> \[\begin{split}f\left(x\right) = \left\{ \begin{array}{lr} \alpha x & : x \lt 0 \\ x & : x \geq 0 \\ \end{array} \right.\\\end{split}\]</div> where alpha is a learned parameter. <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">alpha_initializer (<a class="reference internal" href="../optimization/optimization.html#mxnet.initializer.Initializer" title="mxnet.initializer.Initializer">Initializer</a>) – Initializer for the <cite>embeddings</cite> matrix.</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.ReflectionPad2D"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">ReflectionPad2D</code>(padding=0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/conv_layers.html#ReflectionPad2D">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.ReflectionPad2D" title="Permalink to this definition">¶</a></dt> <dd>Pads the input tensor using the reflection of the input boundary. <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">padding (int) – An integer padding size</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with the shape $(N, C, H_{in}, W_{in})$.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last"> <li>out: output tensor with the shape $(N, C, H_{out}, W_{out})$, where <div class="math"> \[ \begin{align}\begin{aligned}H_{out} = H_{in} + 2 \cdot padding\\W_{out} = W_{in} + 2 \cdot padding\end{aligned}\end{align} \]</div> </li> </ul> </dd> </dl> Examples <div class="highlight-default"><div class="highlight"><pre>>>> m = nn.ReflectionPad2D(3) >>> input = mx.nd.random.normal(shape=(16, 3, 224, 224)) >>> output = m(input) </pre></div> </div> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.SELU"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">SELU</code>(**kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#SELU">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.SELU" title="Permalink to this definition">¶</a></dt> <dd><dl class="docutils"> <dt>Scaled Exponential Linear Unit (SELU)</dt> <dd>“Self-Normalizing Neural Networks”, Klambauer et al, 2017 <a class="reference external" href="https://arxiv.org/abs/1706.02515">https://arxiv.org/abs/1706.02515</a></dd> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </dd></dl> <dl class="class"> <dt id="mxnet.gluon.nn.Swish"> class <code class="descclassname">mxnet.gluon.nn.</code><code class="descname">Swish</code>(beta=1.0, **kwargs)<a class="reference internal" href="../../../_modules/mxnet/gluon/nn/activations.html#Swish">[source]</a><a class="headerlink" href="#mxnet.gluon.nn.Swish" title="Permalink to this definition">¶</a></dt> <dd><dl class="docutils"> <dt>Swish Activation function</dt> <dd><a class="reference external" href="https://arxiv.org/pdf/1710.05941.pdf">https://arxiv.org/pdf/1710.05941.pdf</a></dd> </dl> <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">beta (float) – swish(x) = x * sigmoid(beta*x)</td> </tr> </tbody> </table> <dl class="docutils"> <dt>Inputs:</dt> <dd><ul class="first last simple"> <li>data: input tensor with arbitrary shape.</li> </ul> </dd> <dt>Outputs:</dt> <dd><ul class="first last simple"> <li>out: output tensor with the same shape as <cite>data</cite>.</li> </ul> </dd> </dl> </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 Neural Network Layers</a><ul> <li><a class="reference internal" href="#overview">Overview</a></li> <li><a class="reference internal" href="#basic-layers">Basic Layers</a></li> <li><a class="reference internal" href="#convolutional-layers">Convolutional Layers</a></li> <li><a class="reference internal" href="#pooling-layers">Pooling Layers</a></li> <li><a class="reference internal" href="#activation-layers">Activation Layers</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/1.4.1/api/python/gluon/nn.html (1,905 lines of code) (raw):