Source code for carpedm.util.eval

# Copyright (C) 2018 Neal Digre.
# This software may be modified and distributed under the terms
# of the MIT license. See the LICENSE file for details.

"""Evaluation helpers."""
import tensorflow as tf

[docs]def confusion_matrix(labels, predictions, num_classes): """Generate confusion matrix summary. Args: labels: Ground truth labels. predictions: Predictions. num_classes: Total number of classes. Returns: nothing """ cm = tf.confusion_matrix( tf.reshape(labels, [-1]), tf.reshape(predictions, [-1]), num_classes ) cm = cm / tf.reduce_max(cm) tf.summary.image("confusion", tf.expand_dims(tf.expand_dims(cm, 2), 0))