DAP project, PyTorch, LSTM model
This is for Yuan's Data Analysis Project. I want to implment LSTM-CRF autoencoder in PyTorch.
The data pre-processing part is done.
LSTM in PyTorch training part is done.
Evaluation part is done.
The CRF-LSTM model part is done.
Test this model on CONLL2000, the result shown in below:
Pre-Training part is done. The pre-training dataset can be downloaded from https://nlp.stanford.edu/projects/glove/
The accuracy after pre-training is 92.88%, before pre-training is 90.96%
The marginal decode is done.
The maximum labelwise accuracy part is done.
Split LSTMCRF into two models. Add CRF module.
Add some hand engineer.
The none-pretrain SGD F1 score ~ 88.5
The pretrain SGD F1 score ~ 91.2