search

Accenture / AmpliGraph

Python library for Representation Learning on Knowledge Graphs https://docs.ampligraph.org
Apache License 2.0
2.14k stars 251 forks source link

Early Stopping not working directly #53

Closed PeruBhardwaj closed 5 years ago

PeruBhardwaj commented 5 years ago

Description

If I train the model without early stopping and then with early stopping, it works fine. But on running early stopping directly, trained parameters are not found while loading the model

Actual Behavior

Screen Shot 2019-03-14 at 12 01 51

Expected Behavior

The code should return an array of ranks

Steps to Reproduce

Screen Shot 2019-03-14 at 12 00 25 Screen Shot 2019-03-14 at 12 01 00 Screen Shot 2019-03-14 at 12 01 51
sumitpai commented 5 years ago

Fixed the issue - occurs only when the best value is same as initial value while early stopping i.e. it never found a better model. But FYI, the clone that you are using is very old as we stopped printing the loss per batch long ago. Kindly update the repo.