wenhaofang/RNNDemo

Some demo RNN(Recurrent Neural Network) models implemented with pytorch, including RNN, LSTM, GRU.

RNN Demo

This repository includes some demo RNN(Recurrent Neural Network) models.

Note: The project refers to 动手学深度学习

Datasets:

• dataset1: JayChou Lyrics

Models

• model0: LanguageModel
  • model1: RNNCell
  • model2: LSTMCell
  • model3: GRUCell

Unit Test

• for loader

# loader0: JayChou Lyrics
PYTHONPATH=. python loaders/loader0.py

• for module

# module1: RNNCell
PYTHONPATH=. python modules/module1.py
# module2: LSTMCell
PYTHONPATH=. python modules/module2.py
# module3: GRUCell
PYTHONPATH=. python modules/module3.py

# module0: LanguageModel using RNNCell
PYTHONPATH=. python modules/module0.py --rnn_type rnn
# module0: LanguageModel using LSTMCell
PYTHONPATH=. python modules/module0.py --rnn_type lstm
# module0: LanguageModel using GRUCell
PYTHONPATH=. python modules/module0.py --rnn_type gru

Main Process

• for train

PYTHONPATH=. python main.py --mode train

• for predict

PYTHONPATH=. python main.py --mode predict

You can change the config either in the command line or in the file utils/parser.py

Here are the examples for each module:

# rnn train
PYTHONPATH=. python main.py \
    --mode train \
    --name rnn \
    --rnn_type rnn
# rnn predict
PYTHONPATH=. python main.py \
    --mode predict \
    --name rnn \
    --rnn_type rnn \
    --epoch 250 \
    --prefix 离开

# lstm train
PYTHONPATH=. python main.py \
    --mode train \
    --name lstm \
    --rnn_type lstm
# lstm predict
PYTHONPATH=. python main.py \
    --mode predict \
    --name lstm \
    --rnn_type lstm \
    --epoch 250 \
    --prefix 离开

# gru train
PYTHONPATH=. python main.py \
    --mode train \
    --name gru \
    --rnn_type gru
# gru predict
PYTHONPATH=. python main.py \
    --mode predict \
    --name gru \
    --rnn_type gru \
    --epoch 250 \
    --prefix 离开

Note: You can replace the data files in the data directory to train the model you want.