Boolean mask ops

ndarray/src/main/java/org/tensorflow/ndarray/Shape.java

@@ -275,6 +275,27 @@ public Shape takeLast(int n) {
     return Shape.of(newDimensions);
   }
 
+  /**
+   * Return a {@code end - begin} dimensional shape with dimensions matching this Shape from {@code begin} to {@code end}.
+   * @param begin Where to start the sub-shape.
+   * @param end Where to end the sub-shape, exclusive.
+   * @return the sub-shape bounded by begin and end.
+   */
+  public Shape subShape(int begin, int end){
+    if (end > numDimensions()) {
+      throw new ArrayIndexOutOfBoundsException(
+          "End index " + end + " out of bounds: shape only has " + numDimensions() + " dimensions.");
+    }
+    if (begin < 0) {
+      throw new ArrayIndexOutOfBoundsException(
+          "Begin index " + begin + " out of bounds: cannot be less than 0.");
+    }
+
+    long[] newDimensions = new long[end - begin];
+    System.arraycopy(dimensionSizes, begin, newDimensions, 0, end - begin);
+    return Shape.of(newDimensions);
+  }
+
   /**
    * Returns a new Shape, with a new first dimension added. In order for this call to succeed,
    * {@link Shape#isUnknown()} must be {@code false}.

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

@@ -59,6 +59,8 @@
 import org.tensorflow.op.core.BatchToSpace;
 import org.tensorflow.op.core.BatchToSpaceNd;
 import org.tensorflow.op.core.Bitcast;
+import org.tensorflow.op.core.BooleanMask;
+import org.tensorflow.op.core.BooleanMaskUpdate;
 import org.tensorflow.op.core.BroadcastDynamicShape;
 import org.tensorflow.op.core.BroadcastTo;
 import org.tensorflow.op.core.Bucketize;
@@ -347,10 +349,10 @@ public final class Ops {
 
   public final SignalOps signal;
 
-  public final QuantizationOps quantization;
-
   public final TrainOps train;
 
+  public final QuantizationOps quantization;
+
   private final Scope scope;
 
   private Ops(Scope scope) {
@@ -372,8 +374,8 @@ private Ops(Scope scope) {
     math = new MathOps(this);
     audio = new AudioOps(this);
     signal = new SignalOps(this);
-    quantization = new QuantizationOps(this);
     train = new TrainOps(this);
+    quantization = new QuantizationOps(this);
   }
 
   /**
@@ -989,6 +991,61 @@ public <U extends TType> Bitcast<U> bitcast(Operand<? extends TType> input, Clas
     return Bitcast.create(scope, input, type);
   }
 
+  /**
+   * Apply boolean mask to tensor.  Returns the flat array of each element corresponding to a {@code true} in the mask.
+   *  <p>
+   *  Numpy equivalent is {@code tensor[mask]}.
+   *  <p>
+   *  In general, {@code 0 < dim(mask) = K <= dim(tensor)}, and {@code mask}'s shape must match
+   *  the first K dimensions of {@code tensor}'s shape.  We then have:
+   *    {@code booleanMask(tensor, mask)[i, j1,...,jd] = tensor[i1,...,iK,j1,...,jd]}
+   *  where {@code (i1,...,iK)} is the ith {@code true} entry of {@code mask} (row-major order).
+   *  <p>
+   *  The {@code axis} could be used with {@code mask} to indicate the axis to mask from (it's 0 by default).
+   *  In that case, {@code axis + dim(mask) <= dim(tensor)} and {@code mask}'s shape must match
+   *  the first {@code axis + dim(mask)} dimensions of {@code tensor}'s shape.
+   *
+   * @param scope
+   * @param tensor The tensor to mask.
+   * @param mask The mask to apply.
+   * @param options carries optional attributes values
+   * @return The masked tensor.
+   */
+  public <T extends TType> Operand<T> booleanMask(Operand<T> tensor, Operand<TBool> mask,
+      BooleanMask.Options... options) {
+    return BooleanMask.create(scope, tensor, mask, options);
+  }
+
+  /**
+   * Updates a tensor at the masked values, and returns the updated tensor.  Does not mutate the input tensors.  {@code
+   *  updates} will be broadcasted by default
+   *  <p>
+   *  Numpy equivalent is `tensor[mask] = updates`.
+   *  <p>
+   *  In general, {@code 0 < dim(mask) = K <= dim(tensor)}, and {@code mask}'s shape must match the first K dimensions of
+   *  {@code tensor}'s shape.  We then have: {@code booleanMask(tensor, mask)[i, j1,...,jd] =
+   *  tensor[i1,...,iK,j1,...,jd]} where {@code (i1,...,iK)} is the ith {@code true} entry of {@code mask} (row-major
+   *  order).
+   *  <p>
+   *  The {@code axis} could be used with {@code mask} to indicate the axis to mask from (it's 0 by default). In that
+   *  case, {@code axis + dim(mask) <= dim(tensor)} and {@code mask}'s shape must match the first {@code axis +
+   *  dim(mask)} dimensions of {@code tensor}'s shape.
+   *  <p>
+   *  The shape of {@code updates} should be {@code [n, t_1, t_2, ...]} where {@code n} is the number of true values in
+   *  {@code mask} and {@code t_i} is the {@code i}th dimension of {@code tensor} after {@code axis} and {@code mask}.
+   *  {@code updates} will be broadcasted to this shape by default, which can be disabled using {@code options}.
+   *
+   * @param tensor The tensor to mask.
+   * @param mask The mask to apply.
+   * @param updates the new values
+   * @param options carries optional attributes values
+   * @return The masked tensor.
+   */
+  public <T extends TType> Operand<T> booleanMaskUpdate(Operand<T> tensor, Operand<TBool> mask,
+      Operand<T> updates, BooleanMaskUpdate.Options... options) {
+    return BooleanMaskUpdate.create(scope, tensor, mask, updates, options);
+  }
+
   /**
    * Return the shape of s0 op s1 with broadcast.
    *  <p>
@@ -1834,13 +1891,14 @@ public Constant<TInt32> constant(Shape shape, IntDataBuffer data) {
   }
 
   /**
-   * Creates a scalar of {@code type}, with the value of {@code number}.
-   *  {@code number} may be truncated if it does not fit in the target type.
+   * Creates a scalar of {@code type}, with the value of {@code number}. {@code number} may be truncated if it does not
+   *  fit in the target type.
    *
    * @param type the type of tensor to create.  Must be concrete (i.e. not {@link org.tensorflow.types.family.TFloating})
    * @param number the value of the tensor
    * @return a constant of the passed type
-   * @throws IllegalArgumentException if the type is abstract (i.e. {@link org.tensorflow.types.family.TFloating}) or unknown.
+   * @throws IllegalArgumentException if the type is abstract (i.e. {@link org.tensorflow.types.family.TFloating}) or
+   *  unknown.
    */
   public <T extends TNumber> Constant<T> constant(Class<T> type, Number number) {
     return Constant.tensorOf(scope, type, number);
@@ -1892,14 +1950,14 @@ public <T extends TType> Constant<T> constantOf(T tensor) {
   }
 
   /**
-   * Creates a scalar of the same type as {@code toMatch}, with the value of {@code number}.
-   *  {@code number} may be truncated if it does not fit in the target type.
+   * Creates a scalar of the same type as {@code toMatch}, with the value of {@code number}. {@code number} may be
+   *  truncated if it does not fit in the target type.
    *
    * @param toMatch the operand providing the target type
    * @param number the value of the tensor
    * @return a constant with the same type as {@code toMatch}
-   * @see Ops#constant(Class, Number)
    * @throws IllegalArgumentException if the type is unknown (which should be impossible).
+   * @see Ops#constant(Class, Number)
    */
   public <T extends TNumber> Constant<T> constantOfSameType(Operand<T> toMatch, Number number) {
     return Constant.tensorOfSameType(scope, toMatch, number);

tensorflow-core/tensorflow-core-api/src/main/java/org/tensorflow/op/Scope.java

@@ -125,6 +125,25 @@ public Scope withName(String opName) {
     return new Scope(env, nameScope.withName(opName), controlDependencies, deviceSpec);
   }
 
+  /**
+   * Returns a new scope where added operations will be prefixed by this scope's op name
+   * (set by {@link #withName(String)}), or the given default if it is unset.  This is intended to be used for
+   * composite ops.
+   *
+   * <p>Ops created with this scope will have {@code name/opName/} as the prefix. The actual
+   * name will be unique in the returned scope. All other properties are inherited from the current
+   * scope.
+   *
+   * <p>The default child scope name must match the regular expression {@code [A-Za-z0-9.][A-Za-z0-9_.\-]*}
+   *
+   * @param defaultName name of the sub scope if this scope's name hasn't been set.
+   * @return a new subscope
+   * @throws IllegalArgumentException if the name is invalid
+   */
+  public Scope withNameAsSubScope(String defaultName){
+    return new Scope(env, nameScope.withSubScope(nameScope.makeOpName(defaultName)), controlDependencies, deviceSpec);
+  }
+
   /**
    * Return a new scope that uses the provided device specification for an op.
    *

tensorflow-core/tensorflow-core-api/src/main/java/org/tensorflow/op/core/BooleanMask.java

@@ -0,0 +1,157 @@
+/*
+  Copyright 2021 The TensorFlow Authors. All Rights Reserved.
+
+ Licensed under the Apache License, Version 2.0 (the "License");
+ you may not use this file except in compliance with the License.
+ You may obtain a copy of the License at
+
+     http://www.apache.org/licenses/LICENSE-2.0
+
+ Unless required by applicable law or agreed to in writing, software
+ distributed under the License is distributed on an "AS IS" BASIS,
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ See the License for the specific language governing permissions and
+ limitations under the License.
+ ==============================================================================
+ */
+package org.tensorflow.op.core;
+
+import java.util.Arrays;
+import java.util.Collections;
+import org.tensorflow.Operand;
+import org.tensorflow.ndarray.Shape;
+import org.tensorflow.ndarray.index.Indices;
+import org.tensorflow.op.Scope;
+import org.tensorflow.op.annotation.Endpoint;
+import org.tensorflow.op.annotation.Operator;
+import org.tensorflow.types.TBool;
+import org.tensorflow.types.TInt32;
+import org.tensorflow.types.TInt64;
+import org.tensorflow.types.family.TType;
+
+@Operator
+public abstract class BooleanMask {
+
+  /**
+   * Apply boolean mask to tensor.  Returns the flat array of each element corresponding to a {@code true} in the mask.
+   * <p>
+   * Numpy equivalent is {@code tensor[mask]}.
+   * <p>
+   * In general, {@code 0 < dim(mask) = K <= dim(tensor)}, and {@code mask}'s shape must match
+   * the first K dimensions of {@code tensor}'s shape.  We then have:
+   *   {@code booleanMask(tensor, mask)[i, j1,...,jd] = tensor[i1,...,iK,j1,...,jd]}
+   * where {@code (i1,...,iK)} is the ith {@code true} entry of {@code mask} (row-major order).
+   * <p>
+   * The {@code axis} could be used with {@code mask} to indicate the axis to mask from (it's 0 by default).
+   * In that case, {@code axis + dim(mask) <= dim(tensor)} and {@code mask}'s shape must match
+   * the first {@code axis + dim(mask)} dimensions of {@code tensor}'s shape.
+   *
+   * @param scope
+   * @param tensor The tensor to mask.
+   * @param mask The mask to apply.
+   * @param options carries optional attributes values
+   * @return The masked tensor.
+   */
+  @Endpoint(name = "booleanMask")
+  public static <T extends TType> Operand<T> create(Scope scope, Operand<T> tensor, Operand<TBool> mask,
+      Options... options) {
+
+    scope = scope.withNameAsSubScope("BooleanMask");
+
+    int axis = 0;
+    if (options != null) {
+      for (Options opts : options) {
+        if (opts.axis != null) {
+          axis = opts.axis;
+        }
+      }
+    }
+
+    if (axis < 0) {
+      axis += tensor.rank();
+    }
+
+    Shape maskShape = mask.shape();
+    Shape tensorShape = tensor.shape();
+
+    if (maskShape.numDimensions() == 0) {
+      throw new IllegalArgumentException("Mask cannot be a scalar.");
+    }
+    if (maskShape.hasUnknownDimension()) {
+      throw new IllegalArgumentException("Mask cannot have unknown number of dimensions");
+    }
+
+    Operand<TInt32> axisTensor = Constant.scalarOf(scope, axis);
+    Shape requiredMaskShape = tensorShape.subShape(axis, axis + maskShape.numDimensions());
+    if (!requiredMaskShape.isCompatibleWith(maskShape)) {
+      throw new IllegalArgumentException(
+          "Mask shape " + maskShape + " is not compatible with the required mask shape: " + requiredMaskShape + ".");
+    }
+
+    org.tensorflow.op.core.Shape<TInt32> liveShape = org.tensorflow.op.core.Shape.create(scope, tensor);
+
+    Operand<TInt32> leadingSize = ReduceProd.create(scope,
+        StridedSliceHelper.stridedSlice(scope,
+            liveShape,
+            Indices.range(axis, axis + maskShape.numDimensions())
+        ),
+        Constant.arrayOf(scope, 0)
+    );
+
+    Operand<T> flattened = Reshape.create(scope, tensor, Concat.create(
+        scope,
+        Arrays.asList(
+            StridedSliceHelper.stridedSlice(scope, liveShape, Indices.sliceTo(axis)),
+            Reshape.create(scope, leadingSize, Constant.arrayOf(scope, 1)),
+            StridedSliceHelper.stridedSlice(scope, liveShape, Indices.sliceFrom(axis + maskShape.numDimensions()))
+        ),
+        Constant.scalarOf(scope, 0)
+    ));
+
+    Operand<TBool> flatMask = Reshape.create(scope, mask, Constant.arrayOf(scope, -1));
+
+    Operand<TInt64> indices = Squeeze.create(scope, Where.create(scope, flatMask), Squeeze.axis(Collections.singletonList(1L)));
+    return Gather.create(scope, flattened, indices, axisTensor);
+  }
+
+  /**
+   * Used to indicate the axis to mask from.
+   * {@code axis + dim(mask) <= dim(tensor)} and {@code mask}'s shape must match
+   * the first {@code axis + dim(mask)} dimensions of {@code tensor}'s shape.
+   * @param axis the axis to mask from.  Uses 0 if null.
+   */
+  public static Options axis(Integer axis){
+    return new Options().axis(axis);
+  }
+
+
+  /**
+   * Used to indicate the axis to mask from.
+   * {@code axis + dim(mask) <= dim(tensor)} and {@code mask}'s shape must match
+   * the first {@code axis + dim(mask)} dimensions of {@code tensor}'s shape.
+   * @param axis the axis to mask from.
+   */
+  public static Options axis(int axis){
+    return new Options().axis(axis);
+  }
+
+  /**
+   * Optional attributes for {@link org.tensorflow.op.core.BooleanMask}
+   */
+  public static class Options {
+
+    /**
+     * @param axis (Optional) The axis to mask from, or 0 if not set.
+     */
+    public Options axis(Integer axis) {
+      this.axis = axis;
+      return this;
+    }
+
+    private Integer axis;
+
+    private Options() {
+    }
+  }
+
+}