This is implementation of the paper Do Convolutional Networks need to be Deep for Text Classification?
# for character-level
python3 char_shallownet.py
# for word-level
python3 word_shallownet.py
There are some arguments
| arguments | default | note |
|---|---|---|
| data_dir | '../dataset/' | |
| pos_file | 'rt-polarity.pos' | |
| neg_file | 'rt-polarity.neg' | |
| val_dir | None | |
| val_pos_file | None | |
| val_neg_file | None | |
| model_dir | './model/' | |
| num_class | 2 | |
| num_per_filters | char_shallownet : 700 word_shallownet : 100 |
|
| vocab | 'vocab.pkl' | only in word-level if you set None, automatically download via gluonnlp |
| max_seq_len | char_shallownet : 1014 word_shallownet: None |
|
| batch_size | 128 | |
| seed | 10 | |
| learning_rate | 0.001 | |
| epochs | 1 |
# for character-level
python3 char_densenet.py
# for word-level
python3 word_densenet.py
There are some arguments
| arguments | default | note |
|---|---|---|
| data_dir | '../dataset/' | |
| pos_file | 'rt-polarity.pos' | |
| neg_file | 'rt-polarity.neg' | |
| val_dir | None | |
| val_pos_file | None | |
| val_neg_file | None | |
| model_dir | './model/' | |
| num_class | 2 | |
| vocab | 'vocab.pkl' | only in word-level if you set None, automatically download via gluonnlp |
| max_seq_len | char_shallownet : 1014 word_shallownet: None |
|
| batch_size | 128 | |
| seed | 10 | |
| learning_rate | 0.001 | |
| epochs | 1 |