<p style="text-align: center;"><strong><img alt="" src="https://simg.open-open.com/show/326eaa9a539ec5575022ac39de585d3d.png" /></strong></p>    <p> </p>    <p>TensorFlow 是一个表达机器学习算法的接口，并且是执行算法的实现框架。使用 TensorFlow 表示的计算可以在众多异构的系统上方便地移植，从移动设别如手机或者平板电脑到成千的GPU计算集群上都可以执行。该系统灵活，可以被用来表示很多的算法包括，深度神经网络的训练和推断算法，也已经被用作科研和应用机器学习系统在若干的计算机科学领域或者其他领域中，例如语言识别、计算机视觉、机器人、信息检索、自然语言理解、地理信息抽取和计算药物发现。</p>    <p style="text-align: center;"><a href="https://simg.open-open.com/show/4a67e12961d71d510c83c2aa35a8febb.gif"><img alt="" src="https://simg.open-open.com/show/4a67e12961d71d510c83c2aa35a8febb.gif" /></a></p>    <h2>更新日志</h2>    <ul>     <li>XLA (experimental): initial release of <a href="/misc/goto?guid=4958998680167449824">XLA</a>, a domain-specific compiler for TensorFlow graphs, that targets CPUs and GPUs.</li>     <li>TensorFlow Debugger (tfdbg): command-line interface and API.</li>     <li>New python 3 docker images added.</li>     <li>Made pip packages pypi compliant. TensorFlow can now be installed by <code>pip install tensorflow</code>command.</li>     <li>Several python API calls have been changed to resemble NumPy more closely.</li>     <li>New (experimental) <a href="/misc/goto?guid=4958998878199704727">Java API</a>.</li>     <li>Android: new person detection + tracking demo implementing "Scalable Object Detection using Deep Neural Networks" (with additional YOLO object detector support)</li>     <li>Android: new camera-based image stylization demo based on "A Learned Representation For Artistic Style"</li>     <li>To help you upgrade your existing TensorFlow Python code to match the API changes below, we have prepared a <a href="/misc/goto?guid=4958999221473782555">conversion script</a>.</li>     <li>TensorFlow/models have been moved to a separate github repository.</li>     <li>Division and modulus operators (/, //, %) now match Python (flooring) semantics. This applies to<code>tf.div</code> and <code>tf.mod</code> as well. To obtain forced integer truncation based behaviors you can use<code>tf.truncatediv</code> and <code>tf.truncatemod</code>.</li>     <li><code>tf.divide()</code> is now the recommended division function. <code>tf.div()</code> will remain, but its semantics do not respond to Python 3 or <code>from future</code> mechanisms.</li>     <li>tf.reverse() now takes indices of axes to be reversed. E.g. <code>tf.reverse(a, [True, False, True])</code> must now be written as <code>tf.reverse(a, [0, 2])</code>. <code>tf.reverse_v2()</code> will remain until 1.0 final.</li>     <li><code>tf.mul</code>, <code>tf.sub</code> and <code>tf.neg</code> are deprecated in favor of <code>tf.multiply</code>, <code>tf.subtract</code> and<code>tf.negative</code>.</li>     <li><code>tf.pack</code> and <code>tf.unpack</code> are deprecated in favor of <code>tf.stack</code> and <code>tf.unstack</code>.</li>     <li><code>TensorArray.pack</code> and <code>TensorArray.unpack</code> are getting deprecated in favor of <code>TensorArray.stack</code>and <code>TensorArray.unstack</code>.</li>     <li>The following Python functions have had their arguments changed to use <code>axis</code> when referring to specific dimensions. We have kept the old keyword arguments for compatibility currently, but we will be removing them well before the final 1.0.      <ul>       <li><code>tf.argmax</code>: <code>dimension</code> becomes <code>axis</code></li>       <li><code>tf.argmin</code>: <code>dimension</code> becomes <code>axis</code></li>       <li><code>tf.count_nonzero</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.expand_dims</code>: <code>dim</code> becomes <code>axis</code></li>       <li><code>tf.reduce_all</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_any</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_join</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_logsumexp</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_max</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_mean</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_min</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_prod</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reduce_sum</code>: <code>reduction_indices</code> becomes <code>axis</code></li>       <li><code>tf.reverse_sequence</code>: <code>batch_dim</code> becomes <code>batch_axis</code>, <code>seq_dim</code> becomes <code>seq_axis</code></li>       <li><code>tf.sparse_concat</code>: <code>concat_dim</code> becomes <code>axis</code></li>       <li><code>tf.sparse_reduce_sum</code>: <code>reduction_axes</code> becomes <code>axis</code></li>       <li><code>tf.sparse_reduce_sum_sparse</code>: <code>reduction_axes</code> becomes <code>axis</code></li>       <li><code>tf.sparse_split</code>: <code>split_dim</code> becomes <code>axis</code></li>      </ul> </li>     <li><code>tf.listdiff</code> has been renamed to <code>tf.setdiff1d</code> to match NumPy naming.</li>     <li><code>tf.inv</code> has been renamed to be <code>tf.reciprocal</code> (component-wise reciprocal) to avoid confusion with<code>np.inv</code> which is matrix inversion</li>     <li>tf.round now uses banker's rounding (round to even) semantics to match NumPy.</li>     <li><code>tf.split</code> now takes arguments in a reversed order and with different keywords. In particular, we now match NumPy order as <code>tf.split(value, num_or_size_splits, axis)</code>.</li>     <li><code>tf.sparse_split</code> now takes arguments in reversed order and with different keywords. In particular we now match NumPy order as <code>tf.sparse_split(sp_input, num_split, axis)</code>. NOTE: we have temporarily made <code>tf.sparse_split</code> require keyword arguments.</li>     <li><code>tf.concat</code> now takes arguments in reversed order and with different keywords. In particular we now match NumPy order as <code>tf.concat(values, axis, name)</code>.</li>     <li><code>tf.image.decode_jpeg</code> by default uses the faster DCT method, sacrificing a little fidelity for improved speed. One can revert to the old behavior by specifying the attribute <code>dct_method='INTEGER_ACCURATE'</code>.</li>     <li><code>tf.complex_abs</code> has been removed from the Python interface. <code>tf.abs</code> supports complex tensors and should be used instead.</li>     <li>Template.<code>var_scope</code> property renamed to <code>.variable_scope</code></li>     <li>SyncReplicasOptimizer is removed and SyncReplicasOptimizerV2 renamed to SyncReplicasOptimizer.</li>     <li><code>tf.zeros_initializer()</code> and <code>tf.ones_initializer()</code> now return a callable that must be called with initializer arguments, in your code replace <code>tf.zeros_initializer</code> with <code>tf.zeros_initializer()</code>.</li>     <li><code>SparseTensor.shape</code> has been renamed to <code>SparseTensor.dense_shape</code>. Same for<code>SparseTensorValue.shape</code>.</li>     <li>Replace tf.scalar_summary, tf.histogram_summary, tf.audio_summary, tf.image_summary with tf.summary.scalar, tf.summary.histogram, tf.summary.audio, tf.summary.image, respectively. The new summary ops take name rather than tag as their first argument, meaning summary ops now respect TensorFlow name scopes.</li>     <li>Replace tf.train.SummaryWriter and tf.train.SummaryWriterCache with tf.summary.FileWriter and tf.summary.FileWriterCache.</li>     <li>Removes RegisterShape from public API. Use C++ shape function registration instead.</li>     <li>Deprecated <code>_ref</code> dtypes from the python API.</li>     <li>In the C++ API (in tensorflow/cc), Input, Output, etc. have moved from the tensorflow::ops namespace to tensorflow.</li>     <li>Change arg order for <code>{softmax,sparse_softmax,sigmoid}_cross_entropy_with_logits</code> to be (labels, predictions), and force use of named args.</li>     <li>New op: <code>parallel_stack</code>.</li>     <li>Introducing common tf io compression options constants for RecordReader/RecordWriter.</li>     <li>Add <code>sparse_column_with_vocabulary_file</code>, to specify a feature column that transform string features to IDs, where the mapping is defined by a vocabulary file.</li>     <li>Added <code>index_to_string_table</code> which returns a lookup table that maps indices to strings.</li>     <li>Add <code>string_to_index_table</code>, which returns a lookup table that matches strings to indices.</li>     <li>Add a <code>ParallelForWithWorkerId</code> function.</li>     <li>Add <code>string_to_index_table</code>, which returns a lookup table that matches strings to indices.</li>     <li>Support restore session from checkpoint files in v2 in <code>contrib/session_bundle</code>.</li>     <li>Added a tf.contrib.image.rotate function for arbitrary angles.</li>     <li>Added <code>tf.contrib.framework.filter_variables</code> as a convenience function to filter lists of variables based on regular expressions.</li>     <li><code>make_template()</code> takes an optional <code>custom_getter_ param</code>.</li>     <li>Added comment about how existing directories are handled by <code>recursive_create_dir</code>.</li>     <li>Added an op for QR factorizations.</li>     <li>Divides and mods in Python API now use flooring (Python) semantics.</li>     <li>Android: pre-built libs are now built nightly.</li>     <li>Android: cmake/gradle build for TensorFlow Inference library under <code>contrib/android/cmake</code></li>     <li>Android: Much more robust Session initialization code.</li>     <li>Android: TF stats now exposed directly in demo and log when debug mode is active</li>     <li>Android: new/better README.md documentation</li>     <li>saved_model is available as <code>tf.saved_model</code>.</li>     <li>Empty op is now stateful.</li>     <li>Improve speed of scatter_update on the cpu for ASSIGN operations.</li>     <li>Change <code>reduce_join</code> to treat <code>reduction_indices</code> in the same way as other <code>reduce_</code> ops.</li>     <li>Move <code>TensorForestEstimator</code> to <code>contrib/tensor_forest</code>.</li>     <li>Enable compiler optimizations by default and allow configuration in configure.</li>     <li><code>tf.divide</code> now honors the name field.</li>     <li>Make metrics weight broadcasting more strict.</li>     <li>Add new queue-like <code>StagingArea</code> and new ops: <code>stage</code> and <code>unstage</code>.</li>    </ul>    <h2>下载</h2>    <ul>     <li><a href="/misc/goto?guid=4958999221577460424" rel="nofollow"><strong>Source code</strong> (zip)</a></li>     <li><a href="/misc/goto?guid=4958999221701421207" rel="nofollow"><strong>Source code</strong> (tar.gz)</a></li>    </ul>    <p>本站原创，转载时保留以下信息：<br /> 本文转自：深度开源（open-open.com）<br /> 原文地址：<a href="http://www.open-open.com/news/view/41c75dc0">http://www.open-open.com/news/view/41c75dc0</a></p>