Text generation system based on a mixed corpus of 《毛澤東語錄》(Quotations From Chairman Mao Tse-Tung) and《論語》(Confucian Analects).
| Framework | Model | Optimizer |
|---|---|---|
| PyTorch | RNN (LSTM) | Adam |
$ cd src
$ python3 -m train.train <text-file>
# For more options
$ python3 -m train.train -hOutputs:
model.bin: torch modelcorpus.bin: parsed corpus, mapping, & vocabulary
$ cd src
$ python3 -m generate_text.gen <corpus-bin-file> <model-bin-file>
# For more options
$ python3 -m generate_text.gen -h├── corpus # Raw & parsed corpus
│ ├── corpus.txt # Main corpus file for training
│ ├── luen_yu_clean.txt # Raw corpus with irrelevant words removed
│ ├── luen_yu_raw.txt # Raw corpus
│ ├── luen_yu_sent.txt # Clean corpus seperated into sentences
│ ├── mao_clean.txt # Raw corpus with irrelevant words removed
│ ├── mao_raw.txt # Raw corpus
│ └── mao_sent.txt # Clean corpus seperated into sentences
├── output # Results
│ ├── log # Log files
│ └── model # Pretrained models
│ └── slxx-bsxx-edxx-hdxx-lrxx-drxx-epxx # seq_length, batch_size, embedding_dim, hidden_dim,
│ # learning_rate, dropout, epochs
└── src # Source codes
├── corpus # Corpus processing
│ ├── mix.py # Mix two corpora
│ └── run.sh # Running the script
├── generate_text # Text generation
│ └── gen.py # Text generation
├── train # Model training
│ ├── data.py # Parse data
│ ├── model.py # Main LSTM model
│ └── train.py # Training
├── gen.sh # Running text generation script
└── train.sh # Running training script