Extra Keras Metrics

Additional metrics integrated with the TensorFlow and Keras Neural Network libraries.

How do I install this package?

As usual, just download it using pip:

pip install extra_keras_metrics

How do I use this package?

In addition to importing individual metrics, sets of metrics are also available.

Multi-class metrics

To retrieve an instance of the set of multi-class metrics, use:

from extra_keras_metrics import get_minimal_multiclass_metrics

model = my_keras_model()
model.compile(
    optimizer="nadam",
    loss="categorical_crossentropy",
    metrics=get_minimal_multiclass_metrics()
)

Sparse multi-class metrics

To retrieve an instance of the set of sparse multi-class metrics, use:

from extra_keras_metrics import get_sparse_multiclass_metrics

model = my_keras_model()
model.compile(
    optimizer="nadam",
    loss="sparse_categorical_crossentropy",
    metrics=get_sparse_multiclass_metrics()
)

Note that currently, this only includes categorical accuracy, as it is the only one provided out-of-the-box by TensorFlow. More metrics are planned.

Binary metrics

To retrieve an instance of the set of binary metrics, use:

from extra_keras_metrics import get_standard_binary_metrics

model = my_keras_model()
model.compile(
    optimizer="nadam",
    loss="binary_crossentropy",
    metrics=get_standard_binary_metrics()
)

All the binary metrics

We have implemented a wide range of binary metrics, including some lesser-known ones. To include all available binary metrics, use:

from extra_keras_metrics import get_complete_binary_metrics

model = my_keras_model()
model.compile(
    optimizer="nadam",
    loss="binary_crossentropy",
    metrics=get_complete_binary_metrics()
)

Extras

You might also enjoy these related packages:

• extra_keras_utils - contains commonly used code for Keras projects.
• plot_keras_history - automatically plots Keras training history.