tf1 1.1.1

F1-score metric for TensorFlow

tf1

F1-score metrics for classification models in TensorFlow. There are 3 average modes provided:

• binary
• macro
• micro

Usage

from tf1 import f1_binary

# use f1_binary as any other metric from tf.metrics.*

Note, that due to streaming nature of metric computation process, "macro" and "micro" average metrics should know total number of classes. Use them as follows:

from tf1 import f1_macro, f1_micro

def my_task_f1_macro(
    labels, predictions, num_classes, weights=None,
    metrics_collections=None, updates_collections=None,
    name=None):

    return f1_macro(
        labels=labels,
        predictions=predictions,
        num_classes=123,  # Required
        weights=weights,
        metrics_collections=metrics_collections,
        updates_collections=updates_collections,
        name=name
    )

# use my_task_f1_macro as any other metric from tf.metrics.*