How manthandixit089/we-msdt-former_model works
This repository is a research notebook for training and evaluating an MSDT-Former P300 speller model, including a class-balanced ensemble and target-prior calibration. The documented workflow runs through EEG data loading and preprocessing, training or loading saved checkpoints, then evaluating classification and speller performance. The README identifies the notebook and saved artifacts, but no notebook source was available to verify internal call details; the workflow links below reflect documented behavior only.
Main entry points: nayaModel.ipynb Experiment notebook · model_0.pt Model checkpoints · ensemble_weights.npy Ensemble weights
Components 12
EEG preparation
- Experiment notebook
nayaModel.ipynb - Preprocessing
nayaModel.ipynb
Modeling
- Baseline training
nayaModel.ipynb - Ensemble training
nayaModel.ipynb - Balanced focal loss
nayaModel.ipynb - Model checkpoints
model_0.pt - Ensemble weights
ensemble_weights.npy
Speller evaluation
- Target-prior calibration
nayaModel.ipynb - Character recognition
nayaModel.ipynb - Performance metrics
nayaModel.ipynb
Other components
- Researcher
- Albany dataset
Connections 14
- Researcher → Experiment notebook: runs
- Experiment notebook → Albany dataset: loads data
- Experiment notebook → Preprocessing: prepares epochs
- Preprocessing → Baseline training: provides epochs
- Preprocessing → Ensemble training: provides epochs
- Ensemble training → Balanced focal loss: uses loss
- Experiment notebook → Model checkpoints: loads checkpoints
- Experiment notebook → Ensemble weights: loads weights
- Ensemble training → Model checkpoints: saves models
- Ensemble training → Ensemble weights: saves weights
- Experiment notebook → Target-prior calibration: applies scaling
- Target-prior calibration → Character recognition: supports decoding
- Experiment notebook → Performance metrics: evaluates results
- Character recognition → Performance metrics: reports accuracy