SparseCategoricalCrossentropy

Computes the crossentropy loss between labels and predictions.

Use this crossentropy loss function when there are two or more label classes. The labels are expected to be provided as integers. If you want to provide labels using one-hot representation, please use CategoricalCrossentropy loss. There should be # classes floating point values per feature for predictions and a single floating point value per feature for label.

In the snippet below, there is a single floating point value per example for labels and # classes floating pointing values per example for predictions . The shape of labels is [batch_size] and the shape of predictions is [batch_size, num_classes].

Standalone usage:

    Operand<TFloat32> labels =
        tf.constant(new float[] {1, 2});
    Operand<TFloat32> predictions =
        tf.constant(new float[][] { {0.05f, 0.95f, 0f}, {0.1f, 0.8f, 0.1f} });
    SparseCategoricalCrossentropy sparseCCE = new SparseCategoricalCrossentropy(tf);
    Operand<TFloat32> result = sparseCCE.call(labels, predictions);
    // produces 1.177f
 

Calling with sample weight:

    Operand<TFloat32> sampleWeight = tf.constant(new float[] {0.3f, 0.7f});
    Operand<TFloat32> result = sparseCCE.call(labels, predictions, sampleWeight);
    // produces 0.814f
 

Using SUM reduction type:

    SparseCategoricalCrossentropy sparseCCE = new SparseCategoricalCrossentropy(tf, Reduction.SUM);
    Operand<TFloat32> result = sparseCCE.call(labels, predictions);
    // produces 2.354f
 

Using NONE reduction type:

    SparseCategoricalCrossentropy sparseCCE = new SparseCategoricalCrossentropy(tf, Reduction.NONE);
    Operand<TFloat32> result = sparseCCE.call(labels, predictions);
    // produces [0.0513f, 2.303f]