versions/0.11.0/tutorials/gluon/customop.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="Creating custom operators with numpy" 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="Creating custom operators with numpy" property="og:description"/> <title>Creating custom operators with numpy — 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/0.11.0/searchindex.js"); Search.init();}); </script>          <link href="../../genindex.html" rel="index" title="Index"> <link href="../../search.html" rel="search" title="Search"/> <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/0.11.0/get_started/install.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/0.11.0/tutorials/gluon/gluon.html">About</a></li> <li><a class="main-nav-link" href="https://www.d2l.ai/">Dive into Deep Learning</a></li> <li><a class="main-nav-link" href="https://gluon-cv.mxnet.io">GluonCV Toolkit</a></li> <li><a class="main-nav-link" href="https://gluon-nlp.mxnet.io/">GluonNLP Toolkit</a></li> </ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">API </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu"> <li><a class="main-nav-link" href="/versions/0.11.0/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/scala/index.html">Scala</a></li> </ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">Docs </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu-docs"> <li><a class="main-nav-link" href="/versions/0.11.0/how_to/faq.html">FAQ</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/tutorials/index.html">Tutorials</a> <li><a class="main-nav-link" href="https://github.com/apache/incubator-mxnet/tree/v0.11.0/example">Examples</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/architecture/index.html">Architecture</a></li> <li><a class="main-nav-link" href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home">Developer Wiki</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/model_zoo/index.html">Model Zoo</a></li> <li><a class="main-nav-link" href="https://github.com/onnx/onnx-mxnet">ONNX</a></li> </li></ul> <a aria-expanded="true" aria-haspopup="true" class="main-nav-link dropdown-toggle" data-toggle="dropdown" href="#" role="button">Community </a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu-community"> <li><a class="main-nav-link" href="http://discuss.mxnet.io">Forum</a></li> <li><a class="main-nav-link" href="https://github.com/apache/incubator-mxnet/tree/v0.11.0">Github</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/community/contribute.html">Contribute</a></li> </ul> <a href="#" class="main-nav-link dropdown-toggle" data-toggle="dropdown" role="button" aria-haspopup="true" aria-expanded="true">0.11.0</a><ul id="package-dropdown-menu" class="dropdown-menu"><li><a href="/">master</a></li><li><a href="/versions/1.7.0/">1.7.0</a></li><li><a href=/versions/1.6.0/>1.6.0</a></li><li><a href=/versions/1.5.0/>1.5.0</a></li><li><a href=/versions/1.4.1/>1.4.1</a></li><li><a href=/versions/1.3.1/>1.3.1</a></li><li><a href=/versions/1.2.1/>1.2.1</a></li><li><a href=/versions/1.1.0/>1.1.0</a></li><li><a href=/versions/1.0.0/>1.0.0</a></li><li><a href=/versions/0.12.1/>0.12.1</a></li><li><a href=/versions/0.11.0/>0.11.0</a></li></ul></nav> <script> function getRootPath(){ return "../../" } </script> <div class="burgerIcon dropdown"> <a class="dropdown-toggle" data-toggle="dropdown" href="#" role="button">☰</a> <ul class="dropdown-menu" id="burgerMenu"> <li><a href="/versions/0.11.0/get_started/install.html">Install</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/tutorials/index.html">Tutorials</a></li> <li class="dropdown-submenu dropdown"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">Gluon</a> <ul class="dropdown-menu navbar-menu" id="package-dropdown-menu"> <li><a class="main-nav-link" href="/versions/0.11.0/tutorials/gluon/gluon.html">About</a></li> <li><a class="main-nav-link" href="http://gluon.mxnet.io">The Straight Dope (Tutorials)</a></li> <li><a class="main-nav-link" href="https://gluon-cv.mxnet.io">GluonCV Toolkit</a></li> <li><a class="main-nav-link" href="https://gluon-nlp.mxnet.io/">GluonNLP Toolkit</a></li> </ul> </li> <li class="dropdown-submenu"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">API</a> <ul class="dropdown-menu"> <li><a class="main-nav-link" href="/versions/0.11.0/api/python/index.html">Python</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/c++/index.html">C++</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/julia/index.html">Julia</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/perl/index.html">Perl</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/r/index.html">R</a></li> <li><a class="main-nav-link" href="/versions/0.11.0/api/scala/index.html">Scala</a></li> </ul> </li> <li class="dropdown-submenu"> <a aria-expanded="true" aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" tabindex="-1">Docs</a> <ul class="dropdown-menu"> <li><a href="/versions/0.11.0/how_to/faq.html" tabindex="-1">FAQ</a></li> <li><a href="/versions/0.11.0/tutorials/index.html" tabindex="-1">Tutorials</a></li> <li><a href="https://github.com/apache/incubator-mxnet/tree/v0.11.0/example" tabindex="-1">Examples</a></li> <li><a href="/versions/0.11.0/architecture/index.html" tabindex="-1">Architecture</a></li> <li><a href="https://cwiki.apache.org/confluence/display/MXNET/Apache+MXNet+Home" tabindex="-1">Developer Wiki</a></li> <li><a href="/versions/0.11.0/model_zoo/index.html" tabindex="-1">Gluon Model Zoo</a></li> <li><a href="https://github.com/onnx/onnx-mxnet" tabindex="-1">ONNX</a></li> </ul> </li> <li class="dropdown-submenu dropdown"> <a aria-haspopup="true" class="dropdown-toggle burger-link" data-toggle="dropdown" href="#" role="button" tabindex="-1">Community</a> <ul class="dropdown-menu"> <li><a href="http://discuss.mxnet.io" tabindex="-1">Forum</a></li> <li><a href="https://github.com/apache/incubator-mxnet/tree/v0.11.0" tabindex="-1">Github</a></li> <li><a href="/versions/0.11.0/community/contribute.html" tabindex="-1">Contribute</a></li> </ul> </li> <li id="dropdown-menu-position-anchor-version-mobile" class="dropdown-submenu" style="position: relative"><a href="#" tabindex="-1">0.11.0</a><ul class="dropdown-menu"><li><a tabindex="-1" href=/>master</a></li><li><a tabindex="-1" href=/versions/1.6.0/>1.6.0</a></li><li><a tabindex="-1" href=/versions/1.5.0/>1.5.0</a></li><li><a tabindex="-1" href=/versions/1.4.1/>1.4.1</a></li><li><a tabindex="-1" href=/versions/1.3.1/>1.3.1</a></li><li><a tabindex="-1" href=/versions/1.2.1/>1.2.1</a></li><li><a tabindex="-1" href=/versions/1.1.0/>1.1.0</a></li><li><a tabindex="-1" href=/versions/1.0.0/>1.0.0</a></li><li><a tabindex="-1" href=/versions/0.12.1/>0.12.1</a></li><li><a tabindex="-1" href=/versions/0.11.0/>0.11.0</a></li></ul></li></ul> </div> <div class="plusIcon dropdown"> <a class="dropdown-toggle" data-toggle="dropdown" href="#" role="button"></a> <ul class="dropdown-menu dropdown-menu-right" id="plusMenu"></ul> </div> <div id="search-input-wrap"> <form action="../../search.html" autocomplete="off" class="" method="get" role="search"> <div class="form-group inner-addon left-addon"> <input class="form-control" name="q" placeholder="Search" type="text"/> </div> <input name="check_keywords" type="hidden" value="yes"> <input name="area" type="hidden" value="default"/> </input></form> <div id="search-preview"></div> </div> <div id="searchIcon"> </div>            </div> </div> </div> <script type="text/javascript"> $('body').css('background', 'white'); </script> <div class="container"> <div class="row"> <div aria-label="main navigation" class="sphinxsidebar leftsidebar" role="navigation"> <div class="sphinxsidebarwrapper"> <ul> <li class="toctree-l1"><a class="reference internal" href="../../api/python/index.html">Python Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../api/r/index.html">R Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../api/julia/index.html">Julia Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../api/c++/index.html">C++ Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../api/scala/index.html">Scala Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../api/perl/index.html">Perl Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../how_to/index.html">HowTo Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../../architecture/index.html">System Documents</a></li> <li class="toctree-l1"><a class="reference internal" href="../index.html">Tutorials</a></li> </ul> </div> </div> <div class="content"> <div class="page-tracker"></div> <div class="section" id="creating-custom-operators-with-numpy"> <h1>Creating custom operators with numpy<a class="headerlink" href="#creating-custom-operators-with-numpy" title="Permalink to this headline">¶</a></h1> In this tutorial, we will learn how to build custom operators with numpy in python. We will go through two examples: <ul class="simple"> <li>Custom operator without any <code class="docutils literal">Parameter</code>s</li> <li>Custom operator with <code class="docutils literal">Parameter</code>s</li> </ul> Custom operator in python is easy to develop and good for prototyping, but may hurt performance. If you find it to be a bottleneck, please consider moving to a C++ based implementation in the backend. <div class="highlight-python"><div class="highlight"><pre>import numpy as np import mxnet as mx from mxnet import gluon, autograd </pre></div> </div> <div class="section" id="parameter-less-operators"> <h2>Parameter-less operators<a class="headerlink" href="#parameter-less-operators" title="Permalink to this headline">¶</a></h2> This operator implements the standard sigmoid activation function. This is only for illustration purposes, in real life you would use the built-in operator <code class="docutils literal">mx.nd.relu</code>. <div class="section" id="forward-backward-implementation"> <h3>Forward & backward implementation<a class="headerlink" href="#forward-backward-implementation" title="Permalink to this headline">¶</a></h3> First we implement the forward and backward computation by sub-classing <code class="docutils literal">mx.operator.CustomOp</code>: <div class="highlight-python"><div class="highlight"><pre>class Sigmoid(mx.operator.CustomOp): def forward(self, is_train, req, in_data, out_data, aux): """Implements forward computation. is_train : bool, whether forwarding for training or testing. req : list of {'null', 'write', 'inplace', 'add'}, how to assign to out_data. 'null' means skip assignment, etc. in_data : list of NDArray, input data. out_data : list of NDArray, pre-allocated output buffers. aux : list of NDArray, mutable auxiliary states. Usually not used. """ x = in_data[0].asnumpy() y = 1.0 / (1.0 + np.exp(-x)) self.assign(out_data[0], req[0], mx.nd.array(y)) def backward(self, req, out_grad, in_data, out_data, in_grad, aux): """Implements backward computation req : list of {'null', 'write', 'inplace', 'add'}, how to assign to in_grad out_grad : list of NDArray, gradient w.r.t. output data. in_grad : list of NDArray, gradient w.r.t. input data. This is the output buffer. """ y = out_data[0].asnumpy() dy = out_grad[0].asnumpy() dx = dy*(1.0 - y)*y self.assign(in_grad[0], req[0], mx.nd.array(dx)) </pre></div> </div> </div> <div class="section" id="register-custom-operator"> <h3>Register custom operator<a class="headerlink" href="#register-custom-operator" title="Permalink to this headline">¶</a></h3> Then we need to register the custom op and describe it’s properties like input and output shapes so that mxnet can recognize it. This is done by sub-classing <code class="docutils literal">mx.operator.CustomOpProp</code>: <div class="highlight-python"><div class="highlight"><pre>@mx.operator.register("sigmoid") # register with name "sigmoid" class SigmoidProp(mx.operator.CustomOpProp): def __init__(self): super(SigmoidProp, self).__init__(True) def list_arguments(self): # this can be omitted if you only have 1 input. return ['data'] def list_outputs(self): # this can be omitted if you only have 1 output. return ['output'] def infer_shape(self, in_shapes): """Calculate output shapes from input shapes. This can be omited if all your inputs and outputs have the same shape. in_shapes : list of shape. Shape is described by a tuple of int. """ data_shape = in_shapes[0] output_shape = data_shape # return 3 lists representing inputs shapes, outputs shapes, and aux data shapes. return (data_shape,), (output_shape,), () def create_operator(self, ctx, in_shapes, in_dtypes): # create and return the CustomOp class. return Sigmoid() </pre></div> </div> </div> <div class="section" id="example-usage"> <h3>Example Usage<a class="headerlink" href="#example-usage" title="Permalink to this headline">¶</a></h3> We can now use this operator by calling <code class="docutils literal">mx.nd.Custom</code>: <div class="highlight-python"><div class="highlight"><pre>x = mx.nd.array([0, 1, 2, 3]) # attach gradient buffer to x for autograd x.attach_grad() # forward in a record() section to save computation graph for backward # see autograd tutorial to learn more. with autograd.record(): y = mx.nd.Custom(x, op_type='sigmoid') print(y) </pre></div> </div> <div class="highlight-python"><div class="highlight"><pre># call backward computation y.backward() # gradient is now saved to the grad buffer we attached previously print(x.grad) </pre></div> </div> </div> </div> <div class="section" id="parametrized-operator"> <h2>Parametrized Operator<a class="headerlink" href="#parametrized-operator" title="Permalink to this headline">¶</a></h2> In the second use case we implement an operator with learnable weights. We implement the dense (or fully connected) layer that has one input, one output, and two learnable parameters: weight and bias. The dense operator performs a dot product between data and weight, then add bias to it. <div class="section" id="forward-backward-implementation"> <h3>Forward & backward implementation<a class="headerlink" href="#forward-backward-implementation" title="Permalink to this headline">¶</a></h3> <div class="highlight-python"><div class="highlight"><pre>class Dense(mx.operator.CustomOp): def __init__(self, bias): self._bias = bias def forward(self, is_train, req, in_data, out_data, aux): x = in_data[0].asnumpy() weight = in_data[1].asnumpy() y = x.dot(weight.T) + self._bias self.assign(out_data[0], req[0], mx.nd.array(y)) def backward(self, req, out_grad, in_data, out_data, in_grad, aux): x = in_data[0].asnumpy() dy = out_grad[0].asnumpy() dx = dy.T.dot(x) self.assign(in_grad[0], req[0], mx.nd.array(dx)) </pre></div> </div> </div> <div class="section" id="registration"> <h3>Registration<a class="headerlink" href="#registration" title="Permalink to this headline">¶</a></h3> <div class="highlight-python"><div class="highlight"><pre>@mx.operator.register("dense") # register with name "sigmoid" class DenseProp(mx.operator.CustomOpProp): def __init__(self, bias): super(DenseProp, self).__init__(True) # we use constant bias here to illustrate how to pass arguments # to operators. All arguments are in string format so you need # to convert them back to the type you want. self._bias = float(bias) def list_arguments(self): return ['data', 'weight'] def list_outputs(self): # this can be omitted if you only have 1 output. return ['output'] def infer_shape(self, in_shapes): data_shape = in_shapes[0] weight_shape = in_shapes[1] output_shape = (data_shape[0], weight_shape[0]) # return 3 lists representing inputs shapes, outputs shapes, and aux data shapes. return (data_shape, weight_shape), (output_shape,), () def create_operator(self, ctx, in_shapes, in_dtypes): # create and return the CustomOp class. return Dense(self._bias) </pre></div> </div> </div> <div class="section" id="use-customop-together-with-block"> <h3>Use CustomOp together with Block<a class="headerlink" href="#use-customop-together-with-block" title="Permalink to this headline">¶</a></h3> Parameterized CustomOp are ususally used together with Blocks, which holds the parameter. <div class="highlight-python"><div class="highlight"><pre>class DenseBlock(mx.gluon.Block): def __init__(self, in_channels, channels, bias, **kwargs): super(DenseBlock, self).__init__(**kwargs) self._bias = bias self.weight = self.params.get('weight', shape=(channels, in_channels)) def forward(self, x): ctx = x.context return mx.nd.Custom(x, self.weight.data(ctx), bias=self._bias, op_type='dense') </pre></div> </div> </div> <div class="section" id="example-usage"> <h3>Example usage<a class="headerlink" href="#example-usage" title="Permalink to this headline">¶</a></h3> <div class="highlight-python"><div class="highlight"><pre>dense = DenseBlock(3, 5, 0.1) dense.initialize() x = mx.nd.uniform(shape=(4, 3)) y = dense(x) print(y) </pre></div> </div> </div> </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="#">Creating custom operators with numpy</a><ul> <li><a class="reference internal" href="#parameter-less-operators">Parameter-less operators</a><ul> <li><a class="reference internal" href="#forward-backward-implementation">Forward & backward implementation</a></li> <li><a class="reference internal" href="#register-custom-operator">Register custom operator</a></li> <li><a class="reference internal" href="#example-usage">Example Usage</a></li> </ul> </li> <li><a class="reference internal" href="#parametrized-operator">Parametrized Operator</a><ul> <li><a class="reference internal" href="#forward-backward-implementation">Forward & backward implementation</a></li> <li><a class="reference internal" href="#registration">Registration</a></li> <li><a class="reference internal" href="#use-customop-together-with-block">Use CustomOp together with Block</a></li> <li><a class="reference internal" href="#example-usage">Example usage</a></li> </ul> </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/0.11.0/tutorials/gluon/customop.html (403 lines of code) (raw):