return
def save_epochs_as_txt(epochs, cvdir):
num_ = epochs.events.shape[0]
if not os.path.isdir(cvdir):
os.mkdir(cvdir)
path_events = os.path.join(cvdir, 'events.txt')
np.savetxt(path_events, epochs.events)
data = epochs.get_data()
for j in range(num_):
path_data_ = os.path.join(cvdir, 'data_%d.txt' % j)
np.savetxt(path_data_, data[j])
st = simple_timer()
train = True
fname_list, ortids, event_ids, tmin, t0, tmax = para_setting(train=train)
# show pretty evoked topo
# fname = fname_list[0]
if not os.path.isdir('pics'):
os.mkdir('pics')
for fname in fname_list:
print(fname)
basename = os.path.basename(fname[0:-4])
epochs, raw = get_epochs(fname=fname,
event_id=event_ids,
tmin=tmin,
t0=t0,
tmax=tmax,
def load_train_test(clf,
filedir,
confuse_mat_path,
freq_h=15,
use_good_sensors=False,
decim=10,
n_jobs=12,
scoring='accuracy'):
# initial running timer
st = simple_timer()
# parameters setting
fname_list, ortids, event_ids, tmin, t0, tmax = para_setting(
train=True, filedir=filedir)
# load raw data and epochs
epochs_run = []
for fname in fname_list:
print(fname)
epochs, raw = get_epochs(fname=fname,
event_id=event_ids,
tmin=tmin,
t0=t0,
tmax=tmax,
freq_l=1,
freq_h=freq_h,
decim=decim,
use_good_sensors=use_good_sensors,
get_envlop=False)
epochs_run.append(epochs)
st.click()
num_repeat = 100
num_ort = 6
num_cross = 5
num_timepoint = epochs.get_data().shape[-1]
confuse_mat = np.zeros([
num_ort, num_ort, num_timepoint, num_timepoint, num_cross, num_repeat
])
# stack data
X_all = np.vstack(epochs_run[j].get_data() for j in range(5))
y_all = np.vstack(epochs_run[j].events for j in range(5))[:, 2]
idx_list = np.unique(y_all)
n = len(X_all)
st.click()
for rep_ in range(num_repeat):
# shuffle data
s_ = np.random.permutation(range(n))
X_shuff = X_all.copy()[s_]
y_shuff = y_all.copy()[s_]
# poke data into different orts(referred as idx_)
X_dict = {}
y_dict = {}
for i in range(len(idx_list)):
idx_ = idx_list[i]
tmp = X_shuff[y_shuff == idx_]
X_dict[idx_] = np.vstack(
np.expand_dims(np.mean(tmp[j * 12 + 0:j * 12 + 12], 0), 0)
for j in range(5))
y_dict[idx_] = np.vstack(i + 1 + np.zeros(len(X_dict[idx_])))
# for each combin, seperate train and test data
for combin_ in itertools.combinations(range(len(idx_list)), 2):
combin = list(idx_list[j] for j in combin_)
print(rep_, combin_, combin)
for cross_ in range(5):
cross_train = [0, 1, 2, 3, 4]
cross_train.pop(cross_)
X_train = np.vstack(X_dict[j][cross_train] for j in combin)
y_train = np.ravel(
np.vstack(y_dict[j][cross_train] for j in combin))
X_test = np.vstack(
np.expand_dims(X_dict[j][cross_], 0) for j in combin)
y_test = np.ravel(
np.vstack(
np.expand_dims(y_dict[j][cross_], 0) for j in combin))
# train and test
scores = train_test(X_train,
y_train,
X_test,
y_test,
clf=clf,
scoring=scoring,
n_jobs=n_jobs)
confuse_mat[combin_[0], combin_[1], :, :, cross_,
rep_] = scores
st.click()
np.save(confuse_mat_path, confuse_mat)