/NLP-Assignment-1

Primary LanguagePython

NLP Assigment for RNN, LSTM and GRU

Author: vietph This repo using pytorch using pretrained weights from Glove

How to run:

  1. Install dependences:

    pip install -r requirements.txt

  2. Download the Glove pretrained weights from this drive:

    https://drive.google.com/drive/folders/1u_FuIAfD5Lh8XB337xX8Yy-WTMnhcYlK?usp=sharing

    and replace the ./glove with it.

  3. To run the training:

    python train.py --train_data train/data/dir --model_to_train <RNN/LSTM/GRU> --n_epochs 20 --learning_rate 0.001 --hidden_size 50 --n_epochs 30 --pretrained_size 50

  4. To get the testing results:

    python test.py --model_to_test RNN --hidden_size 50 --pretrained_size 50

Performances on testing dataset:

The performances are recorded on 3 models (RNN|GRU|LSTM) with 3 setting (50d - 50g|100d - 100g|150d - 50g)

d: number of hidden size
g: number of Glove size
RNN GRU LSTM
50d - 50g 0.5921 0.6279 0.6336
100d - 100g 0.6371 0.6386 0.7243
150d - 50g 0.6543 0.6779 0.735

Inference:

Run

python inference.py

There are 3 setting:

  1. 50 hidden_size and 50 Glove size
  2. 100 hidden_size and 100 Glove size
  3. 150 hidden_size and 50 Glove size

Example run:

>>> Choose your model <RNN|LSTM|GRU>: RNN 
>>> Choose the hidden_size <50|100|150>: 100
>>> Choose the glove_size <50|100>:100
Model RNN_model_100_100.pt loaded
Glove loaded
>>> Input sequence or ctrl+d to finish: 
a taut , intelligent psychological drama
Positive