Dareplane c-VEP Decoder

This is a module for the Dareplane project. It provides a decoding module to classify the code-modulated visual evoked potential (c-VEP) from the EEG.

Installation

To download the dp-cvep-speller, use:

git clone https://github.com/thijor/dp-cvep-decoder.git

Make sure that requirements are installed, ideally in a separate conda environment:

conda create --name dp-cvep-decoder python=3.10
conda activate dp-cvep-decoder
pip install -r requirements.txt

Getting started

To run the dp-cvep-speller module in isolation, use:

python -m cvep_decoder.decoder.py

This will run a minimal example using defaults as specified in configs/decoder.toml.

Notes

Be aware of the interplay if the following two timing parameters:

[online]
sleep_s = 0.1                            # time between updates in seconds

[online.eval]
eval_after_type = 'time'                  
eval_after_s = 0.1                        # time after which the model is evaluated

If both are set as in the example above, you might get an unexpected behavior. This is due to the fact that system sleeps are never precise. Which could result in an update cycle producing a prediction and then sleeping for e.g. 90ms to match the target 100ms update cycles defined by sleep_s=0.1. After the sleep, the next update cycle will start. It is possible however, that at this very time, the LSL stream watcher will have collected only samples for 99ms worth of data (the eval_after_s is internally interpreted as a required sample number by: int(eval_after_s * sampling_freq) for the input LSL stream. With these insufficient number of samples, another sleep for about 100ms is triggered. Subsequently, data is processed for about 199ms in the next update cycle, which will then evaluate the classifier. The simplest solution is to have sleep_s, i.e., the update cycle time according to your target, and having eval_after_s smaller than sleep_s. The decoder will then produce decodings in relatively stable intervals of sleep_s. The eval_after_s parameter would be useful, if evaluation should happen much slower than the update cycle (sleep_s).