Precision result different from that obtained with sklearn

[camilomarino]

I wanted to know why when calculating "precision" when the "preds" are of type int I get different results than when using sklearn or doing the math on paper.

Below is a code snippet that shows more clearly what my query is:

import torchmetrics
from sklearn.metrics import precision_score

preds = torch.tensor([0, 0, 1, 0, 0], dtype=torch.int)
target = torch.tensor([0, 0, 1, 1, 1], dtype=torch.int)
print(torchmetrics.functional.precision(preds=preds, target=target))
print(precision_score(y_true=target, y_pred=preds))

preds = torch.tensor([0, 0, 1, 0, 0], dtype=torch.float)
target = torch.tensor([0, 0, 1, 1, 1], dtype=torch.int)
print(torchmetrics.functional.precision(preds=preds, target=target))
print(precision_score(y_true=target, y_pred=preds))

Result:

tensor(0.6000)
1.0
tensor(1.)
1.0

[sraghavm]

Did you rectified the problem

[rjavierch]

Did you find why? I am facing the same issue with regression metrics.

[Borda]

with 1.4 version it seems to be correct:

import torch
from torchmetrics.functional import precision as tm_precision
from sklearn.metrics import precision_score as sklearn_precision

preds = torch.tensor([0, 0, 1, 0, 0], dtype=torch.int)
target = torch.tensor([0, 0, 1, 1, 1], dtype=torch.int)
print("TM: ", tm_precision(preds=preds, target=target, task="binary"))
print("SK: ", sklearn_precision(y_true=target, y_pred=preds))

preds = torch.tensor([0, 0, 1, 0, 0], dtype=torch.float)
target = torch.tensor([0, 0, 1, 1, 1], dtype=torch.int)
print("TM: ", tm_precision(preds=preds, target=target, task="binary"))
print("SK: ", sklearn_precision(y_true=target, y_pred=preds))