Engineering Practice and Scientific and Technological Innovation Homework IV
file name | description |
---|---|
baseline.py |
Basic SVM, ready to run. |
LSTMbaseline.py |
The basic LSTM, running directly will perform training and testing of the network. You can modify the program behavior by passing in parameters, see the parser definition in the code for details. |
TCA.py |
After using TCA for data domain adaptation, use LSTM for classification. Parameters can also be passed in. |
CORAL.py |
Use CORAL's transfer learning method, but the effect is extremely poor, discarded. |
DANN.py |
Build a DANN network for transfer learning. Parameters can also be passed in. |
ADDA.py |
Build an ADDA network for transfer learning. Parameters can also be passed in. |
DANNensemble.py |
Our work, using DANN for ensemble learning and data augmentation. Parameters can also be passed in. |
All models will be saved in the ./saved_models
folder after training. Because the total size of the model generated by 15-fold cross-validation is too large, it is not submitted together here. You can download additional models from here, and unzip it directly to the directory ./
. In networks that allow passing parameters, pass the parameter --predict_only
to skip the training phase and test directly on the test set.
The test results (confusion matrices) of all models are saved in the ./figures
folder. In addition, ./figures/statAll.png
shows a bar chart of the prediction accuracy of all models.