from models.crnl_net import NetAdapter
from skorch_ext.callbacks import LayerDataViz, EpochSummary, GetModuleOnTrainBegin
from functools import partial
from experiments.viz import Viz
viz = Viz()
expe_config = {
'name':
'crnl_net_adam_weighted_filtered',
'dir':
'/dycog/Jeremie/Loic/results',
'datasets': [{
'eeg_dataset': RSVP(),
'include_subjects': ['VPfat', 'VPgcc'],
'exclude_subjects': [],
'apply_filter': {
'l_freq': 0.1,
'h_freq': 20.
},
#epochs params
'exclude_channels': ['P8', 'O2'],
'tmin': 0.0,
'tmax': 0.5,
#others params
'cache': True
}],
'scenario':
import shelve
import glob
import os
import numpy as np
import mne
from experiments.base import ExperimentLogger
from skorch_ext.netsaver import NetSaver
from torch_ext.pytorch_smoothgrad.gradients import VanillaGrad, SmoothGrad
from datasets.rsvp import RSVP
raw_mne = RSVP().get_subject_data('VPgcc').pick_types(eeg=True, exclude=['P8', 'O2'])
def ma(values, window):
weights = np.repeat(1.0, window)/window
sma = np.convolve(values, weights, 'valid')
return sma
if __name__ == '__main__':
import dill as pickle
import copy
d = '/dycog/Jeremie/Loic/results/eeg_net_sgd_weighted_filtered_fat/'
g = os.path.join(d, '*.dat')
for f in glob.glob(g):
experiment_logger = ExperimentLogger(f[:-4], overwrite=False)