This code loads a dataset collected using bela-data-logger and bela-data-syncer and trains a neural network for next sample prediction. For now, it is only possible to train an lstm.
You can create an environment with the necessary dependencies using pipenv:
pipenv install
You will also need to manually install torch:
pipenv run pip3 install torch
and tensorflow if you want to convert the model to tensorflow:
pip env run pip install tensorflow
You can sync data coming from multiple Belas using the script in data/process-data-multi.py
. If you have data coming only from one Bela, you can process it using process-data-single.py
.
Once the dataset is processed, you can modify the necessary paths in train.py
and run the training script by typing:
pipenv run python train.py
The training and model parameters can be modified by passing .yaml
files to the wandb config:
pipenv run python train.py --config configs/test-trans.yaml