Exploiting statistical structure in the data to improve neural decoding

Environment setup

Create conda environment:

conda env create -f env.yaml

Activate the environment:

conda activate decoding

Datasets

Run the following to preprocess and cache IBL datasets:

python src/0_data_caching.py --datasets reproducible-ephys --n_sessions 10 --base_path XXX

Models

We provide example scripts to run the following models:

1. single-session linear / reduced-rank / MLP / LSTM model:

python src/1_decode_single_session.py --eid 5dcee0eb-b34d-4652-acc3-d10afc6eae68 --target choice --method linear --region all --base_path XXX 

2. multi-session reduced-rank model:

python src/2_decode_multi_session.py --target choice --region all --base_path XXX

3. multi-region reduced-rank model:

python src/3_decode_multi_region.py --target choice --query_region CA1 LP PO --base_path XXX

We provide example notebooks to run the following models:

1. single-session / oracle / multi-session BMM-HMM (notebooks/BMM-HMM-example.ipynb)
2. single-session / oracle / multi-session LG-AR1 (notebooks/LG-AR1-example.ipynb)