Initialization in text

[opatrascoiu]

In evaluate method from
https:/explosion/spacy/blob/master/examples/training/train_textcat.py
tp, fp, fn, and tn (true positives etc) are initialized to 1e-8 (see below)

tp = 1e-8  # True positives
fp = 1e-8  # False positives
fn = 1e-8  # False negative
tn = 1e-8  # True negatives

Why are the values not initlaized to 0? When there are no true positives the precision is 0.5, That doesn't seem right.

[honnibal]

Fair point --- I was avoiding the divide-by-zero error, but if there are no true or false positives, I guess the precision should be 0.0. It should be more correct to initialize tp and tn to 0.0, while fp and fn are initialized to a small value to avoid the denominator being zero.

Do you want to make a pull request?

[lock]

This thread has been automatically locked since there has not been any recent activity after it was closed. Please open a new issue for related bugs.