TF Metrics

Multi-class metrics for Tensorflow, similar to scikit-learn multi-class metrics.

Thank you all for making this project live (50-100 clones/day 😎). Contributions welcome!

Install

To add tf_metrics to your current python environment, run

pip install git+https://github.com/guillaumegenthial/tf_metrics.git

For a more advanced use (editable mode, for developers)

git clone https://github.com/guillaumegenthial/tf_metrics.git
cd tf_metrics
pip install -r requirements.txt

Example

Pre-requisite: understand the general tf.metrics API. See for instance the official guide on custom estimators or the official documentation.

Simple example

import tensorflow as tf
import tf_metrics
y_true = [0, 1, 0, 0, 0, 2, 3, 0, 0, 1]
y_pred = [0, 1, 0, 0, 1, 2, 0, 3, 3, 1]
pos_indices = [1, 2, 3]  # Class 0 is the 'negative' class
num_classes = 4
average = 'micro'
# Tuple of (value, update_op)
precision = tf_metrics.precision(
    y_true, y_pred, num_classes, pos_indices, average=average)
recall = tf_metrics.recall(
    y_true, y_pred, num_classes, pos_indices, average=average)
f2 = tf_metrics.fbeta(
    y_true, y_pred, num_classes, pos_indices, average=average, beta=2)
f1 = tf_metrics.f1(
    y_true, y_pred, num_classes, pos_indices, average=average)
# Run the update op and get the updated value
with tf.Session() as sess:
    sess.run(tf.local_variables_initializer())
    sess.run(precision[1])

If you want to use it with tf.estimator.Estimator, add to your model_fn

metrics = {
    'precision': precision,
    'recall': recall,
    'f1': f1,
    'f2': f2
    }
# For Tensorboard
for metric_name, metric in metrics.items():
    tf.summary.scalar(metric_name, metric[1])
if mode == tf.estimator.ModeKeys.EVAL:
    return tf.estimator.EstimatorSpec(
        mode, loss=loss, eval_metric_ops=metrics)