Open kdrivas opened 5 years ago
I wonder about this as well, with the sample config the results are very poor (L2: ~0.1) after training 20 epochs using the WOS dataset.
Hi, I tried to replicate the experiment with the DBpedia dataset as well but the results are very low too even after 50 epochs. Could you give us some guidance ?
2020-02-11 08:56:55,325 - INFO - After Epoch 49
2020-02-11 08:56:55,325 - INFO - Train Loss : 10.252857208251953
2020-02-11 08:56:55,326 - INFO - Train Log Loss : -10.210188865661621
2020-02-11 08:56:55,326 - INFO - Validation Loss : Exact : 10.59323787689209, Overall : 10.59323787689209
2020-02-11 08:56:55,326 - INFO - Train accuracy for level 0 : 0.5625
2020-02-11 08:56:55,327 - INFO - Validation accuracy, Mode: exact, for level 0 : 0.4375
2020-02-11 08:56:55,327 - INFO - Validation accuracy, Mode: overall, for level 0 : 0.4375
2020-02-11 08:56:55,327 - INFO - Validation correct confidence for level 0 : Exact : 0.446709543466568, Overall : 0.446709543466568
2020-02-11 08:56:55,327 - INFO - Validation incorrect confidence for level 0 : Exact : 0.4452267289161682, Overall : 0.4452267289161682
2020-02-11 08:56:55,328 - INFO - Train accuracy for level 1 : 0.125
2020-02-11 08:56:55,328 - INFO - Validation accuracy, Mode: exact, for level 1 : 0.125
2020-02-11 08:56:55,328 - INFO - Validation accuracy, Mode: overall, for level 1 : 0.125
2020-02-11 08:56:55,328 - INFO - Validation correct confidence for level 1 : Exact : 0.0889354795217514, Overall : 0.0889354795217514
2020-02-11 08:56:55,329 - INFO - Validation incorrect confidence for level 1 : Exact : 0.08808653801679611, Overall : 0.08808653801679611
2020-02-11 08:56:55,329 - INFO - Train accuracy for level 2 : 0.0
2020-02-11 08:56:55,329 - INFO - Validation accuracy, Mode: exact, for level 2 : 0.0
2020-02-11 08:56:55,330 - INFO - Validation accuracy, Mode: overall, for level 2 : 0.0
2020-02-11 08:56:55,330 - INFO - Validation correct confidence for level 2 : Exact : 0.0, Overall : 0.0
2020-02-11 08:56:55,330 - INFO - Validation incorrect confidence for level 2 : Exact : 0.012071911245584488, Overall : 0.012071911245584488
Thx in advance
Hi all, first of all thank you for your patience. I apologize I have been overlooking this issue for a while due to other ongoing projects. Currently I'm revamping this codebase with modern best-practices, I will release it shortly with updated information. In the meantime if you notice any irregularities with the codebase feel free to open an issue.
Hi @koustuvsinha, will there be any updates soon? Thx ;-)
I would be grateful if there was a version of keras
Attention at all times
I want to replicate the experiment proposed in your paper, but i think that the file "sample.config.yaml" doesn't contain the original hyperparameters configuration. so could you upload the config file used in the paper?