conda env or requirements.txt file

[bparbhu]

HI guys,

Love the work you're doing here and I'm also looking to enter the competition around this framework. Though one thing
that I'm looking for is a conda env or a requirments.txt file for both models that were trained. I'm not exactly seeing one in either
model that was trained for the neural decoder or the word decoder so I'm just wondering if anyone can provide one or if you can point me in the right direction. Much appreciated and all the best!

-Brian

[cffan]

Hi @bparbhu,

All the python dependencies are in the setup.py. You can create an empty conda environment with python>=3.7, and then do pip install . in NeuralDecoder folder.

[bparbhu]

So I've been getting errors with the setup.py file you have included as the environment you rely on.
The main errors I get are around jupyterlab taking forever to be included in the env when it fetches the package.
The other problem I see is that because of the older version of python you have used for this, some of the releases you have included as dependencies may be deprecated and are not able to be fetched by either conda or pip.

Here are the type of errors I've gotten when omitting jupyterlab.

PackagesNotFoundError: The following packages are not available from current channels:

tensorflow-gpu==2.7.0

[bparbhu]

Also I have included a conda env and a pip env that is made up of the dependencies you have listed and those releases are not able to be found.
requirements.txt

[bparbhu]

name: speechBCI_env
channels:

• defaults
• conda-forge
  dependencies:
• python=3.7 # Adjust the Python version as needed
• tensorflow-gpu=2.7.0
• hydra-core=1.3.2
• hydra-submitit-launcher=1.1.5
• hydra-optuna-sweeper=1.2.0
• transformers=4.28.1
• redis
• seaborn=0.12.2
• pandas
• jupyterlab
• ipywidgets
• tqdm
• g2p_en=2.1.0
• numpy=1.25.0
• scipy=1.11.1
• pytorch=1.13.1 # Note: 'torch' is named 'pytorch' in conda
• accelerate=0.20.3
• bitsandbytes=0.39.1
• edit_distance=1.0.6
• wandb=0.15.5
• hiplot
• numba
• scikit-learn
• pip:
  • bitsandbytes==0.39.1 # Some packages might be available only via pip
  • edit_distance==1.0.6
  • g2p_en==2.1.0
  • wandb==0.15.5
  • accelerate==0.20.3

[bparbhu]

We should address this as it's a barrier to replicating the amazing work done in this paper. We need to arrive at a usable environment that can be used by anyone in the competition.

[cffan]

Thanks for feedbacks!

I can reproduce some of your package missing problems on my end. Sorry about that. I think some dependencies have changed since we wrote the code.

You should be able to install NeuralDecoder with python==3.9. I just tested on my end. If it still doesn't work, please run the following commands and paste the outputs here:

conda create -n bci_test_env python=3.9
conda activate bci_test_env
pip install .