minimum loss is always set to training loss

tensorflow / skflow

Simplified interface for TensorFlow (mimicking Scikit Learn) for Deep Learning

Apache License 2.0

3.18k stars 439 forks source link

minimum loss is always set to training loss #131

Closed mheilman closed 8 years ago

mheilman commented 8 years ago

The code in BaseMonitor.update currently sets the minimum loss so far as follows (link):

        if self.last_loss_seen < self.min_loss:
            self.min_loss = training_loss
            self.min_loss_i = self.steps

I believe the 2nd line there should be self.min_loss = self.last_loss_seen since the validation monitor sets that here.

terrytangyuan commented 8 years ago

@mheilman Thanks for spotting that out! I just pushed the necessary change. You are welcome to submit PR directly to fix issues you found in the future once you signed CLA!

mheilman commented 8 years ago

You're welcome. Thanks for the quick fix. If I happen find anything else in the future, I'll submit a PR (once the CLA stuff is sorted out).