Metrics init scope (#382)

Author: JimClarke5

tensorflow-framework/src/main/java/org/tensorflow/framework/metrics/AUC.java

@@ -16,10 +16,11 @@
 
 import static org.tensorflow.framework.utils.CastHelper.cast;
 
+import org.tensorflow.Graph;
 import org.tensorflow.Operand;
 import org.tensorflow.framework.losses.impl.LossTuple;
 import org.tensorflow.framework.metrics.impl.LossMetric;
-import org.tensorflow.framework.metrics.impl.MeanMetricWrapper;
+import org.tensorflow.framework.metrics.impl.MeanBaseMetricWrapper;
 import org.tensorflow.framework.metrics.impl.MetricsHelper;
 import org.tensorflow.ndarray.Shape;
 import org.tensorflow.op.Ops;
@@ -36,50 +37,54 @@
  *
  * @param <T> The data type for the metric result
  */
-public class Accuracy<T extends TNumber> extends MeanMetricWrapper<T> implements LossMetric<T> {
+public class Accuracy<T extends TNumber> extends MeanBaseMetricWrapper<T> implements LossMetric {
 
   /**
    * Creates an Accuracy Metric using {@link Class#getSimpleName()} for the metric name
    *
-   * @param tf the TensorFlow Ops
    * @param seed the seed for random number generation. An initializer created with a given seed
    *     will always produce the same random tensor for a given shape and data type.
    * @param type the data type for the variables
    */
-  public Accuracy(Ops tf, long seed, Class<T> type) {
-    this(tf, null, seed, type);
+  public Accuracy(long seed, Class<T> type) {
+    this(null, seed, type);
   }
 
   /**
    * Creates an Accuracy Metric
    *
-   * @param tf the TensorFlow Ops
    * @param name the name of the metric, if null then {@link Class#getSimpleName()} is used
    * @param seed the seed for random number generation. An initializer created with a given seed
    *     will always produce the same random tensor for a given shape and data type.
    * @param type the data type for the variables
    */
-  public Accuracy(Ops tf, String name, long seed, Class<T> type) {
-    super(tf, name, seed, type);
+  public Accuracy(String name, long seed, Class<T> type) {
+    super(name, seed, type);
     setLoss(this);
   }
 
   /**
    * Calculates how often predictions equals labels. {@code labels} and {@code predictions} must
    * have compatible shapes, see {@link Shape @isCompatibleWith}.
    *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
    * @param labels the truth values or labels
    * @param predictions the predictions
-   * @throws IllegalArgumentException if predictions and labels shapes are not compatible.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not encapsulate a Graph
+   *     environment.
    * @return the loss
    */
   @Override
-  public Operand<T> call(
-      Operand<? extends TNumber> labels, Operand<? extends TNumber> predictions) {
-    Operand<T> tLabels = cast(getTF(), labels, getResultType());
-    Operand<T> tPredictions = cast(getTF(), predictions, getResultType());
+  public <U extends TNumber> Operand<U> call(
+      Ops tf,
+      Operand<? extends TNumber> labels,
+      Operand<? extends TNumber> predictions,
+      Class<U> resultType) {
+    init(tf);
+    Operand<T> tLabels = cast(tf, labels, getInternalType());
+    Operand<T> tPredictions = cast(tf, predictions, getInternalType());
     LossTuple<T> tuple =
-        MetricsHelper.raggedAssertCompatibleAndGetFlatValues(getTF(), tLabels, tPredictions);
+        MetricsHelper.raggedAssertCompatibleAndGetFlatValues(tf, tLabels, tPredictions);
     tLabels = tuple.getLabels();
     tPredictions = tuple.getTarget();
 
@@ -91,6 +96,6 @@ public Operand<T> call(
     }
 
     // cast TBool to result type
-    return cast(getTF(), getTF().math.equal(tLabels, tPredictions), getResultType());
+    return cast(tf, tf.math.equal(tLabels, tPredictions), resultType);
   }
 }

tensorflow-framework/src/main/java/org/tensorflow/framework/metrics/Accuracy.java

@@ -0,0 +1,259 @@
+/* Copyright 2020 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.framework.metrics;
+
+import java.util.Collections;
+import java.util.List;
+import org.tensorflow.Graph;
+import org.tensorflow.Operand;
+import org.tensorflow.op.Op;
+import org.tensorflow.op.Ops;
+import org.tensorflow.types.family.TNumber;
+
+/** Base class for Metrics */
+public abstract class BaseMetric implements Metric {
+
+  /** The seed for random number generation */
+  private final long seed;
+
+  private String name;
+
+  private boolean initialized;
+
+  private Ops tf;
+
+  /**
+   * Creates a Metric with a name of {@link Class#getSimpleName()}
+   *
+   * @param seed the seed for random number generation. An initializer created with a given seed
+   *     will always produce the same random tensor for a given shape and data type.
+   */
+  protected BaseMetric(long seed) {
+    this(null, seed);
+  }
+
+  /**
+   * Creates a Metric
+   *
+   * @param name the name for this metric. If null, name defaults to {@link Class#getSimpleName()}.
+   * @param seed the seed for random number generation. An initializer created with a given seed
+   *     will always produce the same random tensor for a given shape and data type.
+   */
+  protected BaseMetric(String name, long seed) {
+
+    this.seed = seed;
+    this.name = name != null ? name : this.getClass().getSimpleName();
+  }
+
+  /**
+   * Creates a List of Operations to update the metric state based on input values.
+   *
+   * <p>This is an empty implementation that should be overridden in a subclass, if needed.
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @param values the inputs to be passed to update state, this may not be null
+   * @param sampleWeights sample weights to be applied to the values, may be null.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   * @return a List of Operations to update the metric state
+   */
+  @SuppressWarnings({"unchecked", "unused"})
+  @Override
+  public List<Op> updateStateList(
+      Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights) {
+    checkIsGraph(tf);
+    return Collections.EMPTY_LIST;
+  }
+
+  /**
+   * Creates a List of Operations to update the metric state based on labels and predictions.
+   *
+   * <p>This is an empty implementation that should be overridden in a subclass, if needed.
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @param labels the labels
+   * @param predictions the predictions
+   * @param sampleWeights sample weights to be applied to the metric values, may be null.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   * @return a List of Operations to update the metric state
+   */
+  @Override
+  @SuppressWarnings({"unchecked", "unused"})
+  public List<Op> updateStateList(
+      Ops tf,
+      Operand<? extends TNumber> labels,
+      Operand<? extends TNumber> predictions,
+      Operand<? extends TNumber> sampleWeights) {
+    checkIsGraph(tf);
+    return Collections.EMPTY_LIST;
+  }
+
+  /**
+   * Creates a NoOp Operation with control dependencies to update the metric state
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @param values the inputs to be passed to update state, this may not be null
+   * @param sampleWeights sample weights to be applied to the values, may be null.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   * @return the Operation to update the metric state
+   */
+  public final Op updateState(
+      Ops tf, Operand<? extends TNumber> values, Operand<? extends TNumber> sampleWeights) {
+    checkIsGraph(tf);
+    List<Op> controlOps = updateStateList(tf, values, sampleWeights);
+    return tf.withSubScope("updateState").withControlDependencies(controlOps).noOp();
+  }
+
+  /**
+   * Creates a NoOp Operation with control dependencies to update the metric state
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @param labels the labels
+   * @param predictions the predictions
+   * @param sampleWeights sample weights to be applied to the metric values, may be null.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   * @return the Operation to update the metric state
+   */
+  public final Op updateState(
+      Ops tf,
+      Operand<? extends TNumber> labels,
+      Operand<? extends TNumber> predictions,
+      Operand<? extends TNumber> sampleWeights) {
+    List<Op> controlOps = updateStateList(tf, labels, predictions, sampleWeights);
+    return tf.withSubScope("updateState").withControlDependencies(controlOps).noOp();
+  }
+
+  /**
+   * Calls update state once, followed by a call to get the result
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @param values the inputs to be passed to update state, this may not be null
+   * @param sampleWeights sample weights to be applied to the values, may be null.
+   * @param <T> The data type for the metric result
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   * @return the result, possibly with control dependencies
+   */
+  @Override
+  public final <T extends TNumber> Operand<T> callOnce(
+      Ops tf,
+      Operand<? extends TNumber> values,
+      Operand<? extends TNumber> sampleWeights,
+      Class<T> type) {
+    checkIsGraph(tf);
+    List<Op> controlOps = updateStateList(tf, values, sampleWeights);
+    Ops ltf = tf.withSubScope("callOnce").withControlDependencies(controlOps);
+    return ltf.identity(result(ltf, type));
+  }
+
+  /**
+   * Gets a formatted name for a variable, in the form {@link #name} + "_" + varName.
+   *
+   * @param varName the base name for the variable
+   * @return the formatted variable name
+   */
+  protected String getVariableName(String varName) {
+    return String.format("%s_%s", this.name, varName);
+  }
+
+  /**
+   * The name for this metric. Defaults to {@link Class#getSimpleName()}.
+   *
+   * <p>Gets the name of this metric.
+   *
+   * @return the name of this metric
+   */
+  public String getName() {
+    return name;
+  }
+
+  /**
+   * Sets the metric name
+   *
+   * @param name the metric name
+   */
+  public void setName(String name) {
+    this.name = name;
+  }
+
+  /**
+   * Gets the random number generator seed value
+   *
+   * @return the random number generator seed value
+   */
+  public long getSeed() {
+    return seed;
+  }
+
+  /**
+   * Initialize the TensorFlow Ops
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @throws IllegalArgumentException if the TensorFlow Ops does not have a Graph environment,
+   */
+  protected abstract void init(Ops tf);
+
+  /**
+   * Gets the TensorFlow Ops for this metric
+   *
+   * @return the TensorFlow Ops for this metric.
+   */
+  protected Ops getTF() {
+    return tf;
+  }
+
+  /**
+   * Sets the TensorFlow Ops for this metric.
+   *
+   * <p>This should be set from the {@link #init(Ops)} implementation.
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not have a Graph environment.
+   */
+  protected void setTF(Ops tf) {
+    checkIsGraph(tf);
+    this.tf = tf;
+  }
+
+  /**
+   * Checks whether the Metric is initialized or not.
+   *
+   * @return true if the Metric has been initialized.
+   */
+  public boolean isInitialized() {
+    return initialized;
+  }
+
+  /**
+   * Sets the initialized indicator
+   *
+   * @param initialized the initialized indicator
+   */
+  protected void setInitialized(boolean initialized) {
+    this.initialized = initialized;
+  }
+
+  /**
+   * Checks if the TensorFlow Ops encapsulates a {@link Graph} environment.
+   *
+   * @param tf the TensorFlow Ops encapsulating a {@link Graph} environment.
+   * @throws IllegalArgumentException if the TensorFlow Ops scope does not encapsulate a Graph
+   *     environment.
+   */
+  protected void checkIsGraph(Ops tf) {
+    if (!tf.scope().env().isGraph()) {
+      throw new IllegalArgumentException(
+          "The Ops environment is not a Graph, Graph is required for metrics.");
+    }
+  }
+}