RealTune

This repository serves as a working space for the RealTune project of the ReTune Hackathon 2023.

Mission

The aim is to create a common ReTune-wide interface for real-time decoding of neural signals.

Previous work

Previous work has been done in the following repositories:

• py_neuromodulation, which is already being used to calculate features from neural signals by multiple people of the ReTune community.
• py_neuromodulation also provides some ideas on how to implement decoding and a real-time stream.

Aims

Some of the specific aims of this project are to:

1. Define common use cases
2. Create a flexible interface that can be used across species, tasks, and recording modalities (iEEG, spiking data etc.)
3. Integrate easily with py_neuromodulation
4. Be lightweight (minimal dependencies), fast and easy to use

Example implementation

An example on how the API could look like:

pip install realtune

import time

import numpy as np
import py_neuromodulation as pn
import realtune
import sklearn

sfreq = 1000
model = sklearn.linear_model.LogisticRegression()
processor = pn.DataProcessor(
    sfreq=sfreq,
    settings=settings,
    nm_channels=nm_channels,
)
decoder = realtune.Decoder(model=model)
for i in range(12):
    timestamp = time.time()
    features = processor.process(data=np.random.rand(sfreq, 1))
    label = 0 if i < 5 else 1
    group = i % 2
    decoder.add_features(
        features=features, 
        timestamp=timestamp,
        label=label,
        group=group,
    )
decoder.cross_validate()
decoder.fit_model()
features = processor.process(data=np.random.rand(sfreq, 1))
prediction = decoder.predict(features=features)