MixNet Usage
View source on GitHubTable of contents
Preparation for MixNet’s input
An example of MixNet’s input on the BCIC IV 2a dataset (All settings are set up as optimal settings in the original paper).
Downloading a particular dataset
python download_datasets.py --dataset 'BCIC2a'
Preprocessing based on Spectral Spatial Signals generation over the considered dataset
python prep_spectral_spatial_signals.py --dataset 'BCIC2a'
Build, fit, and evaluate MixNet
# Subject-dependent MI classification
python run_MixNet.py --model_name 'MixNet' --dataset 'BCIC2a' --train_type 'subject_dependent' --data_type 'spectral_spatial_signals' --adaptive_gradient True --policy 'HistoricalTangentSlope' --log_dir 'logs' --num_class 2 --GPU 0 --margin 1.0 --n_component 2 --warmup 7
# Subject-independent MI classification
python run_MixNet.py --model_name 'MixNet' --dataset 'BCIC2a' --train_type 'subject_independent' --data_type 'spectral_spatial_signals' --adaptive_gradient True --policy 'HistoricalTangentSlope' --log_dir 'logs' --num_class 2 --GPU 0 --margin 0.1 --n_component 4 --latent_dim 128 --warmup 5
Arguments:
| Arguments | Description | Default | |
|---|---|---|---|
| model_name | str prefix to save model | ‘MixNet’ | |
| dataset | str prefix to pick up a particular dataset and save model | ‘BCIC2a’ | |
| train_type | str prefix to pick up a particular traning manner and save model | ‘subject_dependent’ | |
| data_type | str prefix to select data type of input | ‘spectral_spatial_signals’ | |
| num_class | int number of classes | 2 | |
| latent_dim | int or None. If None, latent_dim is equal to number of CSP components multiply by nunber of frequency bands U x Nb | None | |
| adaptive_gradient | str2bool to use adaptive loss weights | True | |
| policy | str adaptive gradient policy for MixNet | ‘HistoricalTangentSlope’ | |
| log_dir | str path to save model | ‘logs’ | |
| subjects | int or list of int or None list of range test subject, if None, use all subjects | None | |
| margin | float margin (alpha) of Triplet loss | 1.0 | |
| n_component | int number of CSP components used | None | |
| warmup | int warm-up period of adaptive gradient bleding | 5 | |
| GPU | str select GPU ID | 0 | - |