Took the example code for an LSTM seq2seq model from here (https://keras.io/examples/nlp/lstm_seq2seq/). Added two seq2seq models: One has 2 LSTM cells. The other has 2 LSTM layers with 2 LSTM cells per layer.
To run the scripts, create and activate a virtual environment by running the following command:
python3 -m venv venv;
. ./venv/bin/actgivate;
pip install -r requirements.txt;
To train the plain seq2seq model (no layers or cells), run the following command:
python3 ./scripts/seq2seq_plain.py
This script saves the trained model to 's2s' folder.
To train the seq2seq model with 2 cells (no layers), run the following command:
python3 ./scripts/seq2seq_cells.py
This script saves the trained model to 's2s-cells' folder.
To train the seq2seq model with 2 layers and 2 cells per layer, run the following command:
python3 ./scripts/seq2seq_layers.py
This script saves the trained model to 's2s-layers' folder.
To predict by loading the saved plain seq2seq model (no layers or cells), run the following command:
python3 ./scripts/seq2seq_plain.py predict
To predict by loading the seq2seq model with 2 cells (no layers), run the following command:
python3 ./scripts/seq2seq_cells.py predict
To predict by loading the seq2seq model with 2 layers and 2 cells per layer, run the following command:
python3 ./scripts/seq2seq_layers.py predict