F1 metric for keras

f1_definition.py

import keras.backend as K
def f1(y_true, y_pred):
    true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))
    possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))
    predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))
    precision = true_positives / (predicted_positives + K.epsilon())
    recall = true_positives / (possible_positives + K.epsilon())
    f1_val = 2*(precision*recall)/(precision+recall+K.epsilon())
    return f1_val
  

f1_usage_example.py

# Add "f1" function to "metrics" parameter of "compile" method, example: 
model.compile(optimizer = opt, loss = "categorical_crossentropy", metrics=["accuracy", f1])

# pay attention "f1" is passed as a function not as a string compared to "accuracy".