Andres Suarez
Andres Suarez
Do you have plans to share your pre-trained model? I trained by myself using the WSJ data you provide, but my results, as explained in Issue https://github.com/yaushian/Tree-Transformer/issues/9 are not good....
Could you please share the code to reproduce results in Sect 7.5 of your paper? I am interested in using your model as a masked LM, to then generate sentences...
Inference
@macabdul9 sorry for the large PR, but I had to accumulate improvements to achieve proper inference. Summary of changes: - I further cleaned the Switchboard dataset, to remove all utterances...
If the reminder of the size of the training/validation/test over the batch size is 1. In my usecase, the validation set has 18753 elements, so using a batch size of...
Checkpoints are saved while training, but the implementation doesn't offer an option to restart training from them.
Labels for the train, valid and test datasets are created independently of each other. See https://github.com/macabdul9/CASA-Dialogue-Act-Classifier/blob/051f75d5fd2412a88e0b7d3364ff0ede9fe2ad25/dataset/dataset.py#L17 Even if sorted, they won't be compatible if the validate/test splits don't contain all...
Temporary fix for issue https://github.com/macabdul9/CASA-Dialogue-Act-Classifier/issues/10 It's not very elegant, but it fixes the problem that different data splits use different dictionaries when they don't all contain the same labels
After PR https://github.com/singnet/language-learning/pull/258, the "mini-pipeline" ppln.py that does grammar-learning and parse-evaluation stopped working in my system. I get the following error: ``` Grammar Learner + Tester ppln v.0.0.1 started 2019-10-08...
Currently, the main parse-evaluator doesn't ignore reference sentences with no links when calculating the overall performance metrics (precision, recall, F1). Given this situation, if a reference parse comes with no...
"Content-word" parse-evaluator mode was implemented in "alternative" evaluator in PR: https://github.com/singnet/language-learning/pull/237. It needs to be included in main parse-evaluator at some point, also.