NOTE: this is an old repository with bad code from years ago. Do not use anything as an example.

Requirements

• Ubuntu 16.04 LTS, MacOS High Sierra or newer
• Python 3.6+

How to setup

1. Setup virtualenvironment: virtualenv venv -p python3
2. Activate venv: source venv/bin/activate
3. Install requirements: pip install -r requirements.txt

How to run LSTM + LDA

1. Calculate LDA beta matrix: python calculate_lda.py
2. Train network: python train_lstm.py
3. Generate output text: python lstm_output.py
4. Run analytics on generated text: python analytics.py

You can adjust n_topics and n_cpu_cores in topic_modelling.py.

Network parameters (vocabulary_size, hidden_dim, learning_rate, nepoch) can be adjusted in lstm_model.py

Known issues

If the network is severely under/overfit, it will not encounter a SENTENCE_END token. This makes lstm_output.py hang. Please train a network with preferably at least 100 epochs.