def testcase():
import sys
sessfn = '../../resources/example_data/mindaffectBCI.txt'
from load_mindaffectBCI import load_mindaffectBCI
print("Loading: {}".format(sessfn))
X, Y, coords = load_mindaffectBCI(sessfn, ofs=60)
times = coords[1]['coords']
fs = coords[1]['fs']
ch_names = coords[2]['coords']
print("X({}){}".format([c['name'] for c in coords], X.shape))
print("Y={}".format(Y.shape))
print("fs={}".format(fs))
# visualize the dataset
from analyse_datasets import debug_test_dataset
debug_test_dataset(X,
Y,
coords,
tau_ms=100,
evtlabs=('re', 'fe'),
rank=1,
model='cca')
def datasettest():
# N.B. imports in function to avoid import loop..
from datasets import get_dataset
from model_fitting import MultiCCA, BwdLinearRegression, FwdLinearRegression
from analyse_datasets import debug_test_dataset
from scoreOutput import plot_outputscore
from decodingCurveSupervised import decodingCurveSupervised
if True:
tau_ms = 300
offset_ms = 0
rank = 1
evtlabs = None
l, f, _ = get_dataset('twofinger')
oX, oY, coords = l(f[0])
else:
tau_ms = 20
offset_ms = 0
rank = 8
evtlabs = None
l, f, _ = get_dataset('mark_EMG')
X, Y, coords = l(f[1],
whiten=True,
stopband=((0, 10), (45, 55), (200, -1)),
filterbank=((10, 20), (20, 45), (55, 95), (105, 200)))
oX = X.copy()
oY = Y.copy()
X = oX.copy()
Y = oY.copy()
# test with reduced number classes?
nY = 8
Y = Y[:, :, :nY + 1, :nY]
plt.close('all')
debug_test_dataset(X,
Y,
coords,
tau_ms=tau_ms,
offset_ms=offset_ms,
evtlabs=evtlabs,
rank=8,
outputscore='ip',
model='cca')
fs = coords[1]['fs']
tau = min(X.shape[-2], int(tau_ms * fs / 1000))
offset = int(offset_ms * fs / 1000)
cca = MultiCCA(tau=tau, offset=offset, rank=rank, evtlabs=evtlabs)
cca.fit(X, Y)
Fy = cca.predict(X, Y, dedup0=True)
(_) = decodingCurveSupervised(Fy)
W = cca.W_
R = cca.R_
b = cca.b_
plot_outputscore(X[0, ...], Y[0, ...], W, R)
def testcase():
datadir = '~/data/bci/external_data/dryad/Cocktail Party'
datadir = 'D://external_data/dryad/Cocktail Party'
sessdir = 'EEG/Subject1'
sessfn = ''
X, Y, coords = load_cocktail(datadir, sessdir, sessfn, subtriallen=15)
from analyse_datasets import debug_test_dataset
debug_test_dataset(X, Y, coords, tau_ms=500, rank=4, evtlabs=None)
def testcase():
import sys
from readCapInf import getPosInfo
datadir = '/home/jadref/data/bci/own_experiments/visual/p300_prn_2'
sessdir = ''
sessfn = 'peter/20100714/jf_prep/peter_rc_5_flash.mat'
sessfn = 'peter/20100714/jf_prep/peter_prn_5_flip.mat'
#sessfn = 'alex/20100722/jf_prep/alex_prn_10_flip.mat'
from offline.load_p300_prn import load_p300_prn
X, Y, coords = load_p300_prn(datadir, sessdir, sessfn, ofs=32, stopband=((0,1),(12,-1)), order=6); oX=X.copy(); oY=Y.copy();
fs = coords[1]['fs']
ch_names = coords[2]['coords']
# attach the electrode positions
cnames, xy, xyz, iseeg =getPosInfo(ch_names)
coords[2]['pos2d'] = xy
# set the analysis paramters
tau = int(.6*fs)
evtlabs = ('re','anyre') # ('re', 'ntre') #('re', 'fe') #
times = [i/fs for i in range(tau)]
rank = 10
from analyse_datasets import debug_test_dataset
debug_test_dataset(X, Y, coords, tau_ms=800, evtlabs=evtlabs, rank=rank, regy=.1, regx=.1)
# try with cross-validation
res = cca.cv_fit(X, Y)
Fy=res['estimator']
(_) = decodingCurveSupervised(Fy)
print('FWD')
fwd = FwdLinearRegression(tau=tau, evtlabs=evtlabs)
fwd.fit(X, Y)
print('fwd = {}'.format(fwd))
Fy = fwd.predict(X, Y, dedup0=True)
print("Fy={}".format(Fy.shape))
(_) = decodingCurveSupervised(Fy)
plot_erp(factored2full(fwd.W_, fwd.R_), evtlabs, times, ch_names)
print('BWD')
bwd = BwdLinearRegression(tau=tau, evtlabs=evtlabs)
bwd.fit(X, Y)
print('bwd = {}'.format(bwd))
Fy = bwd.predict(X, Y, dedup0=True)
print("Fy={}".format(Fy.shape))
(_) = decodingCurveSupervised(Fy)
def testcase():
import sys
sessfn = '~/data/bci/own_experiments/emg/facial/training_data_SV_6.mat'
# command-line, for testing
if len(sys.argv) > 1:
sessfn = sys.argv[1]
#from offline.load_mark_EMG import load_mark_EMG
oX, oY, coords = load_mark_EMG(sessfn, ofs=125, stopband=((0,10),(45,55),(95,105),(145,-1)), plot=False)
fs = coords[1]['fs']
ch_names = coords[2]['coords']
X=oX.copy()
Y=oY.copy()
print("X({}){}".format([c['name'] for c in coords],X.shape))
print("Y={}".format(Y.shape))
print("fs={}".format(fs))
# try as simple linear classification problem
from sklearn.linear_model import Ridge, LogisticRegression, LogisticRegressionCV
from sklearn.svm import LinearSVR, LinearSVC
# trial average
Xc = np.sum(X,axis=-2)
Yc = np.argmax(Y[:,0,0,:],axis=-1)
clsfr = LogisticRegression(C=1e8) # turn off regularization
clsfr.fit(Xc,Yc)
Yest = clsfr.predict(Xc)
acc = sum([i==j for i, j in zip(Yc,Yest)]) / len(Yc); print("Acc:{}".format(acc))
W = clsfr.coef_
b = clsfr.intercept_
Fe = np.einsum("Td,ed->Te",Xc,W) + b
Fy = Fe
Yest = np.argmax(Fy,-1)
acc = sum([i==j for i, j in zip(Yc,Yest)]) / len(Yc); print("Acc:{}".format(acc))
# per-sample
clsfr = LogisticRegression(C=1e8)
Xc = X.reshape((-1,X.shape[-1]))
Yc = np.argmax(Y[...,0,:],axis=-1).reshape((-1,))
clsfr.fit(Xc,Yc)
W = clsfr.coef_
b = clsfr.intercept_
Fe = np.einsum("Td,ed->Te",Xc,W) + b
Fy = Fe
Yest = np.argmax(Fy,-1)
acc = sum([i==j for i, j in zip(Yc,Yest)]) / len(Yc); print('Acc: {}'.format(acc))
# visualize the dataset
from analyse_datasets import debug_test_dataset
debug_test_dataset(X, Y, coords, tau_ms=20, evtlabs=None, rank=15, model='cca')
def testcase():
import sys
if os.path.isdir('D:\external_data'):
sessfn = 'D:\external_data\gigadb\openBMI\session1\s1\sml_sess01_subj01_EEG_SSVEP.mat'
sessfn = 'D:\external_data\gigadb\openBMI\session1\s1\sml_sess01_subj01_EEG_ERP.mat'
sessfn = 'D:\external_data\gigadb\openBMI\session1\s2\sml_sess01_subj02_EEG_MI.mat'
else:
sessfn = '/home/jadref/data/bci/external_data/gigadb/openBMI/session1/s2/sml_sess01_subj02_EEG_MI.mat'
# command-line, for testing
if len(sys.argv) > 1:
sessfn = sys.argv[1]
from load_openBMI import load_openBMI
X, Y, coords = load_openBMI(sessfn,
CAR=True,
offset_ms=(-400, 1000),
sessfn=sessfn,
ofs=60,
stopband=((0, 1), (30, -1)))
fs = coords[1]['fs']
# CAR=False,ofs=60,stopband=((0,3),(29,-1)),rcond=1e-3 : audc=36 Perr[-1]=.30
# CAR=True,ofs=60,stopband=((0,3),(29,-1)),rcond=1e-3 : audc=36 Perr[-1]=.30
# CAR=True,ofs=60,stopband=((0,3),(29,-1)),rcond=1e-8 : audc=36 Perr[-1]=.30
if 'SSVEP' in sessfn:
tau_ms = 300
evtlabs = ('re')
rank = 1
elif 'ERP' in sessfn:
tau_ms = 700
evtlabs = ('re', 'ntre')
rank = 10
elif 'MI' in sessfn:
tau_ms = 20
evtlabs = None
rank = 5
# visualize the dataset
from analyse_datasets import debug_test_dataset
debug_test_dataset(X,
Y,
coords,
rank=rank,
evtlabs=evtlabs,
tau_ms=tau_ms,
offset_ms=0,
model='cca',
rcond=1e-2)
def testcase():
try:
from offline.load_ninapro_db2 import load_ninapro_db2
except:
pass
datadir = '/home/jadref/data/bci/external_data/ninapro/DB2_s1/S1_E1_A1.mat'
ofs = 30
X, Y, coords = load_ninapro_db2(datadir,
ofs=ofs,
envelopeband=((0, .01), (5, -1)),
log=True)
from analyse_datasets import debug_test_dataset
debug_test_dataset(X,
Y,
coords,
rank=5,
evtlabs=None,
tau_ms=120,
offset_ms=0,
model='cca')
# get normally pre-processed
X, Y, coords = load_mindaffectBCI(savefile,
stopband=((0, 5), (25, -1)),
order=6,
ofs=200)
# output is: X=eeg, Y=stimulus, coords=meta-info about dimensions of X and Y
print("EEG: X({}){} @{}Hz".format([c['name'] for c in coords], X.shape,
coords[1]['fs']))
print("STIMULUS: Y({}){}".format([c['name'] for c in coords[:1]] + ['output'],
Y.shape))
plt.close('all')
debug_test_dataset(X,
Y,
coords,
tau_ms=350,
evtlabs=('re', 'fe'),
rank=1,
model='cca')
# check the electrode qualities computation
ppfn = butterfilt_and_downsample(
order=6, stopband='butter_stopband((0, 5), (25, -1))_fs200.pk', fs_out=100)
X = X.reshape((-1, X.shape[-1]))
ppfn.fit(X[:10, :], fs=200)
Xp = ppfn.transform(X)
plt.clf()
plt.subplot(211)
plt.plot(X)
plt.subplot(212)
plt.plot(Xp)