Metrics should follow the scikit-learn pattern of accepting two arguments:
percent_positive
(y_true
, y_pred
)
Compute the percent of predictions that are positive. This
can help us identify when a model is predicting all ones or zeros.
mean_soft_prediction
(y_true
, y_score
)
Compute the mean predicted probability.
std_soft_prediction
(y_true
, y_score
)
Compute the standard deviation of the predicted
probabilities. This helps us identify if the model is
always predicting roughly the same probability.
batch_size
(y_true
, y_pred
)
Count the number of items in the current batch.