def extract_words(self, content_word = True, epoch_type = 'epoch', dirty = False):
if not hasattr(self,'raw'):
if dirty: self.load_eeg_data(apply_ica = False)
else: self.load_eeg_data()
if not hasattr(self,'raw') or self.raw == 0:
print('could not load raw, doing nothing')
return
if not hasattr(self,'data'):
if epoch_type in ['epochn400','n400','baseline_n400']:
keep_channels = utils.n400_channel_set()
elif epoch_type in ['epochpmn','pmn','epoch']:
keep_channels = utils.pmn_channel_set()
else: raise ValueError('unk epoch type:',epoch_type)
self.data,self.ch,self.rm_ch= utils.raw2np(self.raw,keep_channels= keep_channels)
self.extracted_words,self.extracted_eeg, self.word_indices = [], [], []
self.epoch_type = epoch_type
for i,w in enumerate(self.words):
if not dirty:
if not w.usable or not hasattr(w,'pos') or not hasattr(w,'ppl'): continue
if content_word and not w.pos.content_word: continue
wd = utils.extract_word_eeg_data(self.data,w, epoch_type = epoch_type)
if type(wd) == np.ndarray:
self.extracted_words.append(w)
self.extracted_eeg.append(wd)
self.word_indices.append(i)
def plot(n_dict, v_dict, gate='190', identifier='cross_entropy1'):
'''plots pmn epochs per channel, not yet tested (reconstructed).
'''
co, cod = cap_order()
values = {}
nd, vd = load_vn_dicts(gate, identifier)
x = np.arange(-300, 1000)
for k in vd.keys():
values[k] = vd[k] / nd[k]
pmn_ch = utils.pmn_channel_set()
for ch in pmn_ch:
rows, cols, index = cod[ch]
plt.subplot(rows, cols, index)
plt.plot(x,
values['low_' + ch] - np.mean(values['low_' + ch][:300]),
color='blue')
plt.plot(x,
values['mid_' + ch] - np.mean(values['mid_' + ch][:300]),
color='orange')
plt.plot(x,
values['high_' + ch] - np.mean(values['high_' + ch][:300]),
color='red')
plt.ylim(0.4, -0.4)
plt.grid()
plt.axis('off')
plt.legend(('low-' + ch, 'mid-' + ch, 'high-' + ch),
fontsize='x-small')
plt.axhline(linewidth=1, color='black')
plt.axvline(linewidth=1, linestyle='-', color='black')
plt.axvline(150, color='tomato', linewidth=1, linestyle='--')
plt.axvline(350, color='tomato', linewidth=1, linestyle='--')
def pmn_dataword2eeg(f, w2e=None):
'''matches eeg data to the pmn dataword and extracts eeg averages for baseline and pmn.
f filename of the pmn_dataword
function is called from make_dataset_gate
'''
if w2e == None: w2e = make_word2eeg_dict()
pmn_ch = utils.pmn_channel_set()
pmnd = open(f).read().rstrip('\n').split('\t')
word_number = f2word_number(f)
failed = []
ch_name2d = {}
output = []
header = load_header()
header = ['pp'] + header
for ch in pmn_ch:
header.extend([ch + '_baseline', ch + '_pmn'])
for i in range(1, 49):
k = str(i) + '_' + word_number
if k not in w2e.keys():
failed.append(k)
continue
fnp = w2e[k]
if not check_match(fnp, pmnd): continue
temp = ['pp' + str(i)] + pmnd[:]
for ch in pmn_ch:
chd = load_eeg_ch(fnp, channel=ch)
if type(chd) == str: temp.extend(['NA', 'NA'])
else:
temp.append(average(chd, 'baseline'))
temp.append(average(chd, 'pmn'))
output.append('\t'.join(map(str, temp)))
return output, '\t'.join(header), failed
def pmn_dataword2ploteeg(f,
content_word,
w2e=None,
n_dict={},
v_dict={},
failed=[],
only_function=False):
'''extract eeg data for a specific pmn_dataword.
function is called from make_plot_gate
'''
if w2e == None: w2e = make_word2eeg_dict()
pmn_ch = utils.pmn_channel_set()
pmnd = open(f).read().rstrip('\n').split('\t')
header = load_header()
if only_function and content_word and pmnd[header.index(
'content_word')] == 'True':
return n_dict, v_dict, failed
elif content_word and pmnd[header.index('content_word')] == 'False':
return n_dict, v_dict, failed
value = float(pmnd[header.index('cross_entropy')])
if value < 0.4: value_type = 'low'
elif 2.7 > value > 0.4: value_type = 'mid'
else: value_type = 'high'
word_number = f2word_number(f)
for i in range(1, 49):
k = str(i) + '_' + word_number
if k not in w2e.keys():
failed.append(k)
continue
fnp = w2e[k]
fch = fnp.replace('.npy', '.ch')
d = np.load(fnp)
channels = open(fch).read().split('\n')
for ch in pmn_ch:
k = value_type + '_' + ch
if k not in n_dict:
n_dict[k] = 0
v_dict[k] = np.zeros(1300)
# chd = load_eeg_ch(fnp,channel=ch)
chd = extract_channel(d, channels, ch)
if type(chd) == str: continue
else:
n_dict[k] += 1
v_dict[k] += chd
return n_dict, v_dict, failed
def pmn_all(content_word=False, w2e=None, only_function=False):
'''Create a plot dict for all gates.'''
if w2e == None: w2e = make_word2eeg_dict()
gates = [
str(g) for g in range(110, 660, 20)
if len(glob.glob(path.pmn_datasets + str(g) + '/WORDS/*')) == 49019
]
values = 'ud_entropy,entropy,cross_entropy,surprisal,ud_surprisal,cow_index_word,ud_index_word'.split(
',')
pmn_ch = utils.pmn_channel_set()
pmnd = open(f).read().rstrip('\n').split('\t')
header = load_header()
if only_function and content_word and pmnd[header.index(
'content_word')] == 'True':
return n_dict, v_dict, failed
elif content_word and pmnd[header.index('content_word')] == 'False':
return n_dict, v_dict, failed
value = float(pmnd[header.index('cross_entropy')])
if value < 1: value_type = 'low'
elif 2.8 > value > 1: value_type = 'mid'
else: value_type = 'high'
word_number = f2word_number(f)
for i in range(1, 49):
k = str(i) + '_' + word_number
if k not in w2e.keys():
failed.append(k)
continue
fnp = w2e[k]
fch = fnp.replace('.npy', '.ch')
d = np.load(fnp)
channels = open(fch).read().split('\n')
for ch in pmn_ch:
k = value_type + '_' + ch
if k not in n_dict:
n_dict[k] = 0
v_dict[k] = np.zeros(1300)
# chd = load_eeg_ch(fnp,channel=ch)
chd = extract_channel(d, channels, ch)
if type(chd) == str: continue
else:
n_dict[k] += 1
v_dict[k] += chd
return n_dict, v_dict, failed
def plot_article(gate='190', identifier='cross_entropy1'):
'''plots pmn epochs per channel, not yet tested (reconstructed).
'''
co, cod = cap_order()
values = {}
nd, vd = load_vn_dicts(gate, identifier)
x = np.arange(-300, 1000)
for k in vd.keys():
values[k] = vd[k] / nd[k]
channels = 'F7,F8,FC5,FC6,T7,T8'.split(',')
rows, cols = 3, 2
pmn_ch = utils.pmn_channel_set()
for ch in channels:
index = channels.index(ch) + 1
plt.subplot(rows, cols, index)
plt.plot(x,
values['low_' + ch] - np.mean(values['low_' + ch][:300]),
color='blue')
plt.plot(x,
values['mid_' + ch] - np.mean(values['mid_' + ch][:300]),
color='orange')
plt.plot(x,
values['high_' + ch] - np.mean(values['high_' + ch][:300]),
color='red')
plt.ylim(0.35, -0.35)
plt.axis('off')
plt.grid()
plt.annotate(ch, xy=(-350, -0.2))
if ch == 'T7':
plt.annotate('-300', xy=(-350, 0.13))
plt.annotate('1000', xy=(900, 0.13))
plt.annotate('-0.3', xy=(-180, -0.3))
plt.annotate(' 0.3', xy=(-180, 0.35))
plt.legend(('low', 'mid', 'high'),
bbox_to_anchor=(1, 1.3),
fontsize='small',
loc=1)
plt.axhline(linewidth=1, color='black')
plt.axvline(linewidth=1, linestyle='-', color='black')
plt.axvline(150, color='tomato', linewidth=1, linestyle='--')
plt.axvline(350, color='tomato', linewidth=1, linestyle='--')