What are all the valid strings I can use with keras.model.compile?

Question

What strings are valid metrics with keras.model.compile?

The following works,

model.compile(optimizer='sgd', loss='mse', metrics=['acc'])

but this does not work,

model.compile(optimizer='sgd', loss='mse', metrics=['recall', 'precision'])

Answer 1

See keras source code, the list of metrics is

• accuracy
• binary_accuracy
• categorical_accuracy
• sparse_categorical_accuracy
• top_k_categorical_accuracy
• sparse_top_k_categorical_accuracy
• cosine_similarity
• binary_crossentropy
• categorical_crossentropy
• categorical_hinge
• hinge
• squared_hinge
• kullback_leibler_divergence
• logcosh
• mean_absolute_error
• mean_absolute_percentage_error
• mean_squared_error
• mean_squared_logarithmic_error
• poisson
• sparse_categorical_crossentropy

Possible abbreviations:

• acc = ACC = accuracy
• bce = BCE = binary_crossentropy
• mse = MSE = mean_squared_error
• mae = MAE = mean_absolute_error
• mape = MAPE = mean_absolute_percentage_error
• msle = MSLE = mean_squared_logarithmic_error
• log_cosh = logcosh
• cosine_proximity = cosine_similarity

any others will need to be added as custom objects, see this link

Answer 2

Check method to check metrices. Check docstring for details