Redo Framework Ops based on the current version of Master. (#342)

Author: JimClarke5

tensorflow-core/tensorflow-core-api/src/bazel/api_def/api_def_SoftmaxCrossEntropyWithLogits.pbtxt

@@ -1,6 +1,6 @@
 op {
   graph_op_name: "SoftmaxCrossEntropyWithLogits"
   endpoint {
-    name: "nn.raw.SoftmaxCrossEntropyWithLogits"
+    name: "nn.SoftmaxCrossEntropyWithLogits"
   }
 }

tensorflow-core/tensorflow-core-api/src/bazel/api_def/api_def_SparseSoftmaxCrossEntropyWithLogits.pbtxt

@@ -1,6 +1,6 @@
 op {
   graph_op_name: "SparseSoftmaxCrossEntropyWithLogits"
   endpoint {
-    name: "nn.raw.SparseSoftmaxCrossEntropyWithLogits"
+    name: "nn.SparseSoftmaxCrossEntropyWithLogits"
   }
 }

tensorflow-core/tensorflow-core-api/src/gen/annotations/org/tensorflow/op/NnOps.java

@@ -83,7 +83,6 @@
 import org.tensorflow.op.nn.Relu;
 import org.tensorflow.op.nn.Relu6;
 import org.tensorflow.op.nn.Selu;
-import org.tensorflow.op.nn.SigmoidCrossEntropyWithLogits;
 import org.tensorflow.op.nn.Softmax;
 import org.tensorflow.op.nn.SoftmaxCrossEntropyWithLogits;
 import org.tensorflow.op.nn.Softsign;
@@ -103,16 +102,13 @@
  * @see {@link Ops}
  */
 public final class NnOps {
-  public final NnRawOps raw;
-
   private final Scope scope;
 
   private final Ops ops;
 
   NnOps(Ops ops) {
     this.scope = ops.scope();
     this.ops = ops;
-    raw = new NnRawOps(ops);
   }
 
   /**
@@ -1815,55 +1811,6 @@ public <T extends TNumber> Selu<T> selu(Operand<T> features) {
     return Selu.create(scope, features);
   }
 
-  /**
-   * Computes sigmoid cross entropy given <code>logits</code>.
-   *
-   *  <p>Measures the probability error in discrete classification tasks in which each class is
-   *  independent and not mutually exclusive. For instance, one could perform multilabel
-   *  classification where a picture can contain both an elephant and a dog at the same time.
-   *
-   *  <p>For brevity, let <code>x = logits</code>, <code>z = labels</code>. The logistic loss in
-   *  pseudo-code is
-   *
-   *  <pre>
-   *  z * -log(sigmoid(x)) + (1 - z) * -log(1 - sigmoid(x))
-   *   = z * -log(1 / (1 + exp(-x))) + (1 - z) * -log(exp(-x) / (1 + exp(-x)))
-   *   = z * log(1 + exp(-x)) + (1 - z) * (-log(exp(-x)) + log(1 + exp(-x)))
-   *   = z * log(1 + exp(-x)) + (1 - z) * (x + log(1 + exp(-x))
-   *   = (1 - z) * x + log(1 + exp(-x))
-   *   = x - x * z + log(1 + exp(-x))
-   *  </pre>
-   *
-   *  <p>For <code>x < 0</code>, to avoid overflow in <code>exp(-x)</code>, we reformulate the above
-   *
-   *  <pre>
-   *  x - x * z + log(1 + exp(-x))
-   *   = log(exp(x)) - x * z + log(1 + exp(-x))
-   *   = - x * z + log(1 + exp(x))
-   *  </pre>
-   *
-   *  <p>Hence, to ensure stability and avoid overflow, the implementation uses this equivalent
-   *  formulation
-   *
-   *  <pre>
-   *    max(x, 0) - x * z + log(1 + exp(-abs(x)))
-   *  </pre>
-   *
-   *  <p></ode>logits</code> and <code>labels</code> must have the same type and shape.
-   *
-   *  <p>
-   *
-   * @param labels the labels
-   * @param logits the logits of type float32 or float64
-   * @param <T> the type of labels and logits
-   * @return the component-wise logistic losses.
-   * @throws IllegalArgumentException if logits' and labels' do not have the same shape
-   */
-  public <T extends TNumber> Operand<T> sigmoidCrossEntropyWithLogits(Operand<T> labels,
-      Operand<T> logits) {
-    return SigmoidCrossEntropyWithLogits.sigmoidCrossEntropyWithLogits(scope, labels, logits);
-  }
-
   /**
    * Computes softmax activations.
    *  For each batch {@code i} and class {@code j} we have
@@ -1881,53 +1828,20 @@ public <T extends TNumber> Softmax<T> softmax(Operand<T> logits) {
   }
 
   /**
-   * Computes softmax cross entropy between <code>logits</code> and <code>labels</code>.
-   *
-   *  <p>Measures the probability error in discrete classification tasks in which the classes are
-   *  mutually exclusive (each entry is in exactly one class). For example, each CIFAR-10 image is
-   *  labeled with one and only one label: an image can be a dog or a truck, but not both.
-   *
-   *  <p><b>NOTE:</b>
-   *
-   *  <p>While the classes are mutually exclusive, their probabilities need not be. All that is
-   *  required is that each row of <code>labels</code> is a valid probability distribution. If they
-   *  are not, the computation of the gradient will be incorrect.
-   *
-   *  <p>If using exclusive <code>labels</code> (wherein one and only one class is true at a time),
-   *  see {@link org.tensorflow.op.NnOps#sparseSoftmaxCrossEntropyWithLogits}
-   *
-   *  <p>Usage:
-   *
-   *  <pre>
-   *    Operand&lt;TFloat32&gt; logits =
-   *        tf.constant(new float[][] {{4.0F, 2.0F, 1.0F}, {0.0F, 5.0F, 1.0F}} );
-   *    Operand&lt;TFloat32&gt; labels =
-   *        tf.constant(new float[][] {{1.0F, 0.0F, 0.0F}, {0.0F, 0.8F, 0.2F}} );
-   *    Operand&lt;TFloat32&gt; output =
-   *        tf.nn.softmaxCrossEntropyWithLogits(labels, logits, -1);
-   *    // output Shape = [2]
-   *    // dataType = FLOAT (1)
-   *    // values { 0.169846, 0.824745 }
-   *  </pre>
-   *
-   *  <p>Backpropagation will happen into both <code>logits</code> and <code>labels</code>. To
-   *  disallow backpropagation into <code>labels</code>, pass label tensors through <code>
-   *  tf.stopGradient</code> before feeding it to this function.
+   * Computes softmax cross entropy cost and gradients to backpropagate.
+   *  Inputs are the logits, not probabilities.
    *
-   * @param labels Each vector along the class dimension should hold a valid probability
-   *      distribution e.g. for the case in which labels are of shape <code>[batch_size, num_classes]
-   *      </code>, each row of <code>labels[i]</code> must be a valid probability distribution.
-   * @param logits Per-label activations, typically a linear output. These activation energies are
-   *      interpreted as unnormalized log probabilities.
-   * @param axis The class dimension. -1 is the last dimension.
-   * @param <T> the number type of the operands
-   * @return the softmax cross entropy loss. Its type is the same as <code>logits</code> and its
-   *      shape is the same as <code>labels</code> except that it does not have the last dimension of
-   *      <code>labels</code>.
+   * @param <T> data type for {@code loss} output
+   * @param features batch_size x num_classes matrix
+   * @param labels batch_size x num_classes matrix
+   *  The caller must ensure that each batch of labels represents a valid
+   *  probability distribution.
+   * @param <T> data type for {@code SoftmaxCrossEntropyWithLogits} output and operands
+   * @return a new instance of SoftmaxCrossEntropyWithLogits
    */
-  public <T extends TNumber, U extends TNumber> Operand<T> softmaxCrossEntropyWithLogits(
-      Operand<U> labels, Operand<T> logits, int axis) {
-    return SoftmaxCrossEntropyWithLogits.softmaxCrossEntropyWithLogits(scope, labels, logits, axis);
+  public <T extends TNumber> SoftmaxCrossEntropyWithLogits<T> softmaxCrossEntropyWithLogits(
+      Operand<T> features, Operand<T> labels) {
+    return SoftmaxCrossEntropyWithLogits.create(scope, features, labels);
   }
 
   /**
@@ -2114,50 +2028,23 @@ public <T extends TType> SpaceToDepth<T> spaceToDepth(Operand<T> input, Long blo
   }
 
   /**
-   * Computes sparse softmax cross entropy between <code>logits</code> and <code>labels</code>.
-   *
-   *  <p>Measures the probability error in discrete classification tasks in which the classes are
-   *  mutually exclusive (each entry is in exactly one class). For example, each CIFAR-10 image is
-   *  labeled with one and only one label: an image can be a dog or a truck, but not both.
-   *
-   *  <p><b>NOTE:</b>
-   *
-   *  <p>For this operation, the probability of a given label is considered exclusive. That is, soft
-   *  classes are not allowed, and the <code>labels</code> vector must provide a single specific
-   *  index for the true class for each row of <code>logits</code> (each minibatch entry). For soft
-   *  softmax classification with a probability distribution for each entry, {@link
-   *  org.tensorflow.op.NnOps#softmaxCrossEntropyWithLogits}.
-   *
-   *  <p><b>WARNING:</b>
+   * Computes softmax cross entropy cost and gradients to backpropagate.
+   *  Unlike {@code SoftmaxCrossEntropyWithLogits}, this operation does not accept
+   *  a matrix of label probabilities, but rather a single label per row
+   *  of features.  This label is considered to have probability 1.0 for the
+   *  given row.
+   *  <p>Inputs are the logits, not probabilities.
    *
-   *  <p>This op expects unscaled logits, since it performs a <code>softmax</code> on <code>logits
-   *  </code> internally for efficiency. Do not call this op with the output of <code>softmax</code>,
-   *  as it will produce incorrect results.
-   *
-   *  <p>A common use case is to have logits of shape <code>[batchSize, numClasses]</code> and have
-   *  labels of shape <code>[batchSize]</code>, but higher dimensions are supported, in which case
-   *  the <code>dim</code>-th dimension is assumed to be of size <code>numClasses</code>. <code>
-   *  logits</code> must have the <cod>dataType</cod> of <code>TFloat16</code>, <code>TFloat32</code>
-   *  , or <code>TFloat64</code>, and <code>labels</code> must have the dtype of <code>TInt32</code>
-   *  or <code>TInt64</code>.
-   *
-   * @param labels <code>Tensor</code> of shape <code>[d_0, d_1, ..., d_{r-1}]</code> (where <code>r
-   *      </code> is rank of <code>labels</code> and result) and the dataType is <code>TInt32</code>
-   *      or <code>TInt64</code>. Each entry in <code>labels</code> must be an index in <code>[0,
-   *      numClasses)</code>. Other values will raise an exception when this op is run on CPU, and
-   *      return <code>NaN</code> for corresponding loss and gradient rows on GPU.
-   * @param logits Per-label activations (typically a linear output) of shape <code>[d_0, d_1, ...,
-   *      d_{r-1}, numClasses]</code> and dataType of <code>TFloat16</code>, <code>TFloat32</code>,
-   *      or <code>TFloat64</code>. These activation energies are interpreted as unnormalized log
-   *      probabilities.
-   * @return A <code>Tensor</code> of the same shape as <code>labels</code> and of the same type as
-   *      <code>logits</code> with the softmax cross entropy loss.
-   * @throws IllegalArgumentException If logits are scalars (need to have rank >= 1) or if the rank
-   *      of the labels is not equal to the rank of the logits minus one.
-   */
-  public <T extends TNumber, U extends TNumber> Operand sparseSoftmaxCrossEntropyWithLogits(
-      Operand<T> labels, Operand<U> logits) {
-    return SparseSoftmaxCrossEntropyWithLogits.sparseSoftmaxCrossEntropyWithLogits(scope, labels, logits);
+   * @param <T> data type for {@code loss} output
+   * @param features batch_size x num_classes matrix
+   * @param labels batch_size vector with values in [0, num_classes).
+   *  This is the label for the given minibatch entry.
+   * @param <T> data type for {@code SparseSoftmaxCrossEntropyWithLogits} output and operands
+   * @return a new instance of SparseSoftmaxCrossEntropyWithLogits
+   */
+  public <T extends TNumber> SparseSoftmaxCrossEntropyWithLogits<T> sparseSoftmaxCrossEntropyWithLogits(
+      Operand<T> features, Operand<? extends TNumber> labels) {
+    return SparseSoftmaxCrossEntropyWithLogits.create(scope, features, labels);
   }
 
   /**

tensorflow-core/tensorflow-core-api/src/gen/annotations/org/tensorflow/op/NnRawOps.java

@@ -367,10 +367,10 @@ public final class Ops {
 
   public final SignalOps signal;
 
-  public final TrainOps train;
-
   public final QuantizationOps quantization;
 
+  public final TrainOps train;
+
   private final Scope scope;
 
   private Ops(Scope scope) {
@@ -393,8 +393,8 @@ private Ops(Scope scope) {
     math = new MathOps(this);
     audio = new AudioOps(this);
     signal = new SignalOps(this);
-    train = new TrainOps(this);
     quantization = new QuantizationOps(this);
+    train = new TrainOps(this);
   }
 
   /**

tensorflow-core/tensorflow-core-api/src/gen/annotations/org/tensorflow/op/Ops.java

@@ -3150,7 +3150,7 @@ public static native void TF_RegisterFilesystemPlugin(
 // TF_InitKernel to do op/kernel registration.
 // Plugin should implement TF_InitKernel to register kernels. This function
 // should register all kernels in a plugin.
-public static native void TF_InitKernel();
+
 // Targeting ../Create_func_TF_OpKernelConstruction.java
 
 

tensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/internal/c_api/global/tensorflow.java

@@ -15,7 +15,7 @@
 
 // This class has been generated, DO NOT EDIT!
 
-package org.tensorflow.op.nn.raw;
+package org.tensorflow.op.nn;
 
 import org.tensorflow.Operand;
 import org.tensorflow.Operation;
@@ -34,7 +34,7 @@
  * @param <T> data type for {@code loss} output
  */
 @Operator(
-    group = "nn.raw"
+    group = "nn"
 )
 public final class SoftmaxCrossEntropyWithLogits<T extends TNumber> extends RawOp {
   /**
@@ -50,7 +50,7 @@ private SoftmaxCrossEntropyWithLogits(Operation operation) {
     super(operation);
     int outputIdx = 0;
     loss = operation.output(outputIdx++);
-    backprop = operation.output(outputIdx++);
+    backprop = operation.output(outputIdx);
   }
 
   /**

…n/raw/SoftmaxCrossEntropyWithLogits.java …op/nn/SoftmaxCrossEntropyWithLogits.javatensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/op/nn/raw/SoftmaxCrossEntropyWithLogits.java renamed to tensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/op/nn/SoftmaxCrossEntropyWithLogits.java tensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/op/nn/raw/SoftmaxCrossEntropyWithLogits.java renamed to tensorflow-core/tensorflow-core-api/src/gen/java/org/tensorflow/op/nn/SoftmaxCrossEntropyWithLogits.java

@@ -15,7 +15,7 @@
 
 // This class has been generated, DO NOT EDIT!
 
-package org.tensorflow.op.nn.raw;
+package org.tensorflow.op.nn;
 
 import org.tensorflow.Operand;
 import org.tensorflow.Operation;
@@ -38,7 +38,7 @@
  * @param <T> data type for {@code loss} output
  */
 @Operator(
-    group = "nn.raw"
+    group = "nn"
 )
 public final class SparseSoftmaxCrossEntropyWithLogits<T extends TNumber> extends RawOp {
   /**
@@ -54,7 +54,7 @@ private SparseSoftmaxCrossEntropyWithLogits(Operation operation) {
     super(operation);
     int outputIdx = 0;
     loss = operation.output(outputIdx++);
-    backprop = operation.output(outputIdx++);
+    backprop = operation.output(outputIdx);
   }
 
   /**