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

Runtime error for early stopping with FB15k-237 #60

Closed PeruBhardwaj closed 5 years ago

PeruBhardwaj commented 5 years ago

Description

I am trying to replicate the performance results from the documentation but getting a runtime error when using early stopping with FB15k-237.

I have used early stopping with other datasets like FB15k, WN18 and there is no runtime error with these.

Steps to Reproduce

Screen Shot 2019-03-20 at 17 15 43 Screen Shot 2019-03-20 at 17 17 05 Screen Shot 2019-03-20 at 17 18 09
sumitpai commented 5 years ago

fb15k-237 and wn18rr datasets have some entities in test and validation set that are not in train set. So we need to remove them before fitting/evaluation. That's why you are getting this error.

PeruBhardwaj commented 5 years ago

Will the train_test_split_no_unseen() function work for this?

sumitpai commented 5 years ago

Just checked. There is a clean_data method in predictive_performance.py which can be used. You can also refer that file for more details.

I think we must move the clean_data to model_utils. I will create a separate issue for that.

sumitpai commented 5 years ago

Closing this issue and creating a separate issue for moving clean_data to a meaningful location