This file realized DNN Query Classification based on DNN and LSTM RNN. Some features in this program: * Using LSTM RNN to understand the context of a specific word in a sentence * Using skip layers to learn the linear relations between regular features and output *
- Train Query Size:46000+
- Test Query Size:6000+
* max query length:20
* bach size: 10
* dropout 0.3
##Structure and Layers * Input (1300)—> LSTM RNN layer —> Fully Connected layer A * Reg Expression (1200)—> Fully Connected layer A * Input (1300) —> Skip Layer A * Reg Expression (1200) —> Skip Layer B * Fully Connected Layer A +Skip Layer A+ Skip Layer B—> Output Layer
##Result Training Accuray : 98% Test Accuracy : 93.7%
| Method | Test Accuracy |
|---|---|
| One NN | 90% |
| Two NNs | 91.3% |
| LSTM+NN | 93.5% |
| GRU +NN | 93.7 |
##Files Explanation *data_io_lstm.py: manipulating raw data, generating training data and test data *train_lstm.py: Initialize the entire program *train_dnn_lstm.py: Initialize loading data, call DNN to fit and train model *dnn_lstm.py: All the model structure, training and fitting process