Julius/semesterproject_lecture_eeg
1
0
Fork 0
Code Issues Pull Requests Releases Activity

Edited some comments

Browse Source
This commit is contained in:
Julius Voggesberger
2021-03-28 15:45:14 +02:00
parent bb3406bd88
commit cfa950b85b
5 changed files with 45 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

34
decoding_tf_analysis.py
View File

@@ -46,6 +46,7 @@ def events_to_labels(evts, events_dict, mask=None): # TODO Test schreiben
def permutation_test(baseline, score, n_iter):
"""
An implementation of a permutation test for classification scores.
:param baseline: The classification scores of the baseline, i.e. selection by chance
:param score: The classification scores which are tested for significance
:param n_iter: number of permutations
@@ -120,6 +121,8 @@ def decoding(dataset, filename, compute_metric=True, mask=None):
# Compute index of time point 0
index = math.floor((len(metric[0]) / time_scale) * 100)
baseline = np.array(metric[:index]).flatten()
# Plot the result
plt.plot(np.linspace(-200, 1000, 1127), np.mean(metric, axis=0))
plt.ylabel('Accuracy (%)')
plt.xlabel('Time (ms)')
@@ -129,11 +132,12 @@ def decoding(dataset, filename, compute_metric=True, mask=None):
# Compute the permutation tests
for t in range(len(metric[0][index:])):
score_t = np.asarray(metric[:, t + index])
p = permutation_test(baseline, score_t, 100)
p = permutation_test(baseline, score_t, 1000)
p_values.append(p)
if t % 50 == 0:
print(str(t) + " Out of " + str(len(metric[0][index:])))
# Plot the result
plt.plot(times[index:], p_values)
plt.ylabel('P-Value')
plt.xlabel('Time (ms)')
@@ -143,14 +147,17 @@ def decoding(dataset, filename, compute_metric=True, mask=None):
def create_tfr(raw, condition, freqs, n_cycles, response='induced', baseline=None, plot=False):
"""
Compute the time frequency representation (TFR) of data for a given condition via morlet wavelets
Compute the time frequency representation (TFR) of data for a given condition via Morlet wavelets
:param raw: the data
:param condition: the condition for which to compute the TFR. Given as a list of tuples of the form (stimulus, condition) # TODO ambiguous use of condition
:param condition: the condition for which to compute the TFR. Given as a list of tuples of the form (stimulus, texture)
:param freqs: the frequencies for which to compute the TFR
:param n_cycles: the number of cycles used by the morlet wavelets
:param response: type of expected TFR. Can be total, induced or evoked. Default is induced
:param n_cycles: the number of cycles used by the Morlet wavelets
:param response: type of expected TFR. Can be total, induced or evoked. Default is induced,
the others were not used for the report, only for exploration
:param baseline: baseline used to correct the power. A tuple of the form (start, end).
Default is None and no baseline correction will be applid
Default is None and no baseline correction will be applied
:param plot: True if results should be plotted, else false.
:return: The TFR or the given data for a given condition. Has type AverageTFR
"""
epochs, _ = get_epochs(raw, condition, tmin=-0.2, tmax=1)
@@ -168,6 +175,7 @@ def create_tfr(raw, condition, freqs, n_cycles, response='induced', baseline=Non
power_induced = tfr_morlet(epochs.subtract_evoked(), freqs=freqs, n_cycles=n_cycles, return_itc=False, n_jobs=4)
power = mne.combine_evoked([power_total, power_induced], weights=[1, -1])
if plot: power.plot(picks='P7')
# Apply a baseline correction to the power data
power.apply_baseline(mode='ratio', baseline=baseline)
if plot:
plot_oscillation_bands(power)
@@ -185,13 +193,9 @@ def time_frequency(dataset, filename, compute_tfr=True):
:param compute_tfr: If True the TFRs will be created, else the TFRs will be loaded from a precomputed file
"""
# Parameters
# Frequency space (from, to, steps) -> Control frequency resolution : Between num=50-80 good for 1-50Hz
# freqs = np.linspace(0.1, 50, num=50) # Use this for linear space scaling
freqs = np.logspace(*np.log10([0.1, 50]), num=50)
# Number of cycles -> Controls time resolution ? At ~freqs/2 good for high frequency resolution
n_cycles = freqs / 2 # 1 for high time resolution & freq smoothing, freqs/2 for high freq resolution & time smooth
# Baseline -> Should not go post-stimulus, i.e. > 0 -> Best ist pre-stimulus (e.g. -400 to -200ms)
baseline = [-0.5, 0]
n_cycles = freqs / 2
cond1 = []
cond2 = []
times = None
@@ -209,6 +213,8 @@ def time_frequency(dataset, filename, compute_tfr=True):
raw.set_montage('standard_1020', match_case=False)
# Create the two conditions we want to compare
# IMPORTANT: If different conditions should be compared you have to change them here, by altering the second
# argument passed to create_tfr
power_cond1 = create_tfr(raw, [('face', 'intact')], freqs, n_cycles, 'induced', (-0.2, 0))
print(' CONDITION 1 LOADED')
cond1.append(power_cond1)
@@ -219,17 +225,23 @@ def time_frequency(dataset, filename, compute_tfr=True):
cond2.append(power_cond2)
print(' DONE')
# Save the data so we can access the results more easily
np.save('cached_data/tf_data/' + filename + '_cond1', cond1)
np.save('cached_data/tf_data/' + filename + '_cond2', cond2)
else:
# If the data should not be recomputed, load the given filename
cond1 = np.load('cached_data/tf_data/' + filename + '_cond1.npy', allow_pickle=True).tolist()
cond2 = np.load('cached_data/tf_data/' + filename + '_cond2.npy', allow_pickle=True).tolist()
if times is None:
times = cond1[0].times
# Some plots
mne.grand_average(cond1).plot(picks=['P7'], vmin=-3, vmax=3, title='Grand Average P7')
mne.grand_average(cond2).plot(picks=['P7'], vmin=-3, vmax=3, title='Grand Average P7')
plot_oscillation_bands(mne.grand_average(cond1))
plot_oscillation_bands(mne.grand_average(cond2))
# Compute the cluster permutation
F, clusters, cluster_p_values, h0 = mne.stats.permutation_cluster_test(
[mne.grand_average(cond1).data, mne.grand_average(cond2).data], n_jobs=4, verbose='INFO',
seed=123)