-
Notifications
You must be signed in to change notification settings - Fork 0
/
run_supplement.py
331 lines (270 loc) · 10.5 KB
/
run_supplement.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Thu Mar 9 10:54:53 2023
code to produce the supplementary figures
@author: simon.kern
"""
import json
import compress_pickle
import matplotlib.pyplot as plt
import mne
import numpy as np
import pandas as pd
import seaborn as sns
from joblib import Parallel, delayed
from tqdm import tqdm
import settings
import utils
from load_funcs import load_localizers_seq12, load_testing_img2
clf = utils.LogisticRegressionOvaNegX(C=6, penalty="l1", neg_x_ratio=2)
files = utils.list_files(settings.data_dir, patterns=["*DSMR*"])
subjects = [f"DSMR{subj}" for subj in sorted(set(map(utils.get_id, files)))]
results_dir = settings.results_dir + "/reactivation/"
pkl_dir = settings.cache_dir + "/pkl"
# load data, will be needed for some of the plots
localizer = {} # localizer training data
testing_img2 = {} # testing trials after RS2
for subj in tqdm(subjects, desc="Loading data"):
localizer[subj] = load_localizers_seq12(subj=subj)
testing_img2[subj] = load_testing_img2(subj=subj, tmin=-0.1, tmax=0.5)
plt.rc("font", size=12)
input("press enter to continue plotting")
#%% S1: rejected trials per participant
json_files = utils.list_files(settings.results_dir + "/log/")
df_rejected = pd.DataFrame()
# loop over all json files, these contain the logs for preprocessing
for file in json_files:
subj = f"DSMR{utils.get_id(file)}"
with open(file, "r") as f:
c = json.load(f)
# first localizer
name = "autoreject_epochs event_ids={1, 2, 3, 4, 5, 6, 7, 8, 9, 10} n_events=240"
perc_bad = sum([[c[f"localizer{i}"][name]["perc_bad"]][0] for i in [1, 2]]) / 2
df_tmp = pd.DataFrame(
{"subject": subj, "ratio bad": perc_bad, "type": "localizer"}, index=[0]
)
df_rejected = pd.concat([df_rejected, df_tmp])
# then testing
name = "autoreject_epochs event_ids={1, 2, 3, 4, 5, 6, 7, 8, 9, 10} n_events=12"
test = "test" if "test" in c else "testing" # I sometimes named it differently
perc_bad = [c[test][name]["perc_bad"]][0]
df_tmp = pd.DataFrame(
{"subject": subj, "ratio bad": perc_bad, "type": "test"}, index=[0]
)
df_rejected = pd.concat([df_rejected, df_tmp], ignore_index=True)
# plot as bar graph
fig = plt.figure(figsize=[12, 8])
ax = fig.subplots(1, 1)
sns.barplot(data=df_rejected, x="subject", y="ratio bad", hue="type")
ax.tick_params(axis="x", rotation=70)
ax.set_ylim([0, 1])
plt.pause(0.1)
fig.tight_layout()
fig.savefig(results_dir + "/S1 - rejected epochs", bbox_inches="tight")
#%% S2: participant's memory performance and rejection
json_files = utils.list_files(settings.results_dir + "/log/")
df_rejected = pd.DataFrame()
# retrieve memory performance
performance = [utils.get_performance(subj=subj) for subj in subjects]
# retrieve decoding accuracy
accuracies = [utils.get_decoding_accuracy(subj=subj, clf=clf)[0] for subj in subjects]
# plot as bar graph
fig = plt.figure(figsize=[12, 8])
ax1, ax2 = fig.subplots(2, 1)
sns.despine()
df_perf_acc = pd.DataFrame(
{
"accuracy": [np.mean(accuracies), 0] + accuracies,
"performance": [np.mean(performance), 0] + performance,
"subject": ["Mean", ""] + subjects,
}
)
palette = ["darkblue", "blue"] + [
"royalblue" if acc >= 0.3 else "grey" for acc in accuracies
]
sns.barplot(data=df_perf_acc, x="subject", y="accuracy", ax=ax1, palette=palette)
ax1.hlines(0.3, 0, 32, linestyle="--", color="gray")
ax1.tick_params(axis="x", rotation=70)
palette = ["darkred", "blue"] + [
"indianred" if perf >= 0.5 else "grey" for perf in performance
]
sns.barplot(data=df_perf_acc, x="subject", y="performance", ax=ax2, palette=palette)
ax2.hlines(0.5, 0, 32, linestyle="--", color="gray")
ax2.tick_params(axis="x", rotation=70)
plt.pause(0.1)
fig.tight_layout()
fig.savefig(results_dir + "/S2 - performance and accuracy.png", bbox_inches="tight")
#%% S3: decoding accuracy over time per participant
# load precomputed localizer decoding accuracies
pkl_localizer = pkl_dir + "/1 localizer.pkl.zip"
df_localizer = pd.read_pickle(pkl_localizer)
fig = plt.figure()
axs = fig.subplots(6, 5)
axs = axs.flatten()
for i, (subj, df_subj) in enumerate(tqdm(df_localizer.groupby("subject"), total=30)):
utils.plot_decoding_accuracy(
df_subj,
x="timepoint",
y="accuracy",
title=f"{subj}",
color="tab:blue",
ax=axs[i],
)
plt.pause(0.1)
fig.tight_layout()
fig.savefig(results_dir + "/S3 - localizer performance.png", bbox_inches="tight")
#%% S4 histogram: number of learning blocks per participant
from utils import get_performance
n_blocks = [len(get_performance(subj=s, which="learning")) for s in subjects]
fig = plt.figure(figsize=[6, 4])
ax = fig.subplots(1, 1)
sns.histplot(
data=pd.DataFrame({"subject": subjects, "n blocks": n_blocks}),
x="n blocks",
bins=np.arange(1, 7) + 0.5,
)
fig.suptitle("Number of learning blocks per participant")
fig.tight_layout
fig.savefig(results_dir + "/S8 learning blocks.svg")
fig.savefig(results_dir + "/S8 learning blocks.png")
from mne.decoding import SlidingEstimator
#%% S5 sensor map: L1 beta coefficient > 0 visualization per class
import meg_tools
def fit(clf, subj, *args, **kwargs):
np.random.seed(utils.get_id(subj))
return clf.fit(*args, **kwargs)
clfs = Parallel(len(localizer) - 1)(
delayed(fit)(clf, subj, X=localizer[subj][0][:, :, 31], y=localizer[subj][1])
for subj in localizer
)
# these are the orders of images as they were assigned to the classes
names = [utils.get_image_names(subj) for subj in subjects]
names_base = sorted(names[0])
# we need to sort the rows of the matrix accordingly
sorting = [[names_base.index(x) for x in name] for name in names]
betas = np.array([clf.coef_[idxs, :] for clf, idxs in zip(clfs, sorting)])
fig = plt.figure(figsize=[10, 7])
axs = fig.subplots(3, 4)
axs = axs.flatten()
for i in range(10):
meg_tools.plot_sensors(
np.mean(betas > 0, 0)[i], ax=axs[i], mode="size", title=names_base[i]
)
axs[-1].axis("off")
axs[-2].axis("off")
fig.suptitle("Sensor distribution for different images")
plt.pause(0.1)
fig.tight_layout
fig.savefig(results_dir + "/S6 sensorlocation.svg")
fig.savefig(results_dir + "/S6 sensorlocation.png")
#%% SX confusion matrix: decoder class balance/decodability per class
# did not make it into the paper
from sklearn.metrics import confusion_matrix
pkl_localizer = pkl_dir + "/1 localizer.pkl.zip"
df_localizer = pd.read_pickle(pkl_localizer)
df_localizer = df_localizer[df_localizer.timepoint == 210]
# these are the orders of images as they were assigned to the classes
names = [utils.get_image_names(subj) for subj in subjects]
names_base = sorted(names[0])
# we need to sort the rows of the matrix accordingly
sorting = [[names_base.index(x) for x in name] for name in names]
confmats = []
for (subj, df_subj), labels in zip(df_localizer.groupby("subject"), sorting):
preds = np.hstack(df_subj.preds)
truth = np.hstack([np.arange(10)] * len(df_subj))
confmat = confusion_matrix(truth, preds, labels=labels, normalize="true")
confmats.append(confmat)
fig = plt.figure(figsize=[6, 5])
ax = fig.subplots(1, 1)
sns.heatmap(np.mean(confmats, 0), ax=ax)
ax.set_xticks(np.arange(10) + 0.5, names_base, rotation=45)
ax.set_yticks(np.arange(10) + 0.5, names_base, rotation=0)
ax.set_xlabel("item name")
ax.set_ylabel("item name")
fig.suptitle("Confusion matrix of decoded items during localizer")
plt.pause(0.1)
fig.tight_layout()
fig.savefig(results_dir + "/S5 confmat.svg")
fig.savefig(results_dir + "/S5 confmat.png")
#%% SX decoding probabilities per class
# did not make it to the final paper
from sklearn.model_selection import StratifiedKFold
ex_per_fold = 1
# these are the orders of images as they were assigned to the classes
items = {subj: utils.get_image_names(subj) for subj in subjects}
items_base = sorted(list(items.values())[0])
# we need to sort the rows of the matrix accordingly
sorting = {subj: [items_base.index(x) for x in item] for subj, item in items.items()}
clf_x = SlidingEstimator(clf)
def train_predict_proba(clf, data_x, data_y, best_tp):
cv = StratifiedKFold(np.bincount(data_y)[0])
probas_current = []
probas_others = []
def inverse_index(i, n=10):
idx = np.ones(n, dtype=bool)
idx[i] = False
return idx
for idx_train, idx_test in cv.split(data_x, data_y):
x_train = data_x[idx_train]
y_train = data_y[idx_train]
x_test = data_x[idx_test]
y_test = data_y[idx_test]
clf.fit(x_train[:, :, best_tp], y_train)
proba = np.array(
[clf.predict_proba(x_test[:, :, t]) for t in range(x_test.shape[-1])]
)
proba = np.swapaxes(proba, 0, 1)
proba_currrent = np.array([p[:, y] for p, y in zip(proba, y_test, strict=True)])
proba_other = np.mean(
[p[:, inverse_index(y)] for p, y in zip(proba, y_test)], -1
)
probas_current.append(proba_currrent)
probas_others.append(proba_other)
probas_current = np.vstack(probas_current)
probas_others = np.vstack(probas_others)
return probas_current, probas_others
times = np.arange(-100, 510, 10)
# calculate in parallel
res = Parallel(10)(
delayed(train_predict_proba)(clf, *localizer[subj], 31) for subj in tqdm(subjects)
)
df_proba = pd.DataFrame()
for i, (probas_current, probas_others) in enumerate(tqdm(res, desc="subject")):
subj = subjects[i]
data_x, data_y = localizer[subj]
assert len(data_x) == len(probas_current)
df_subj1 = pd.DataFrame(
{
"probability": probas_current.ravel(),
"timepoint": np.hstack([times] * len(probas_current)),
"image": np.repeat(
[items_base[sorting[subj][y]] for y in data_y], probas_current.shape[-1]
),
"subject": subj,
"stimulus": "current",
}
)
df_subj2 = pd.DataFrame(
{
"probability": probas_others.ravel(),
"timepoint": np.hstack([times] * len(probas_others)),
"image": np.repeat(
[items_base[sorting[subj][y]] for y in data_y], probas_others.shape[-1]
),
"subject": subj,
"stimulus": "other",
}
)
df_proba = pd.concat([df_proba, df_subj1, df_subj2])
fig, ax = plt.subplots(1, 1, figsize=[6, 6])
sns.lineplot(
data=df_proba, x="timepoint", y="probability", hue="stimulus", style="image", ax=ax
)
sns.despine()
ax.grid("on")
ax.set_title("raw class probabilities")
ax.set_xlabel("ms after stimulus onset")
fig.tight_layout()
fig.savefig(results_dir + "/localizer_raw_probabilities.png")
fig.savefig(results_dir + "/localizer_raw_probabilities.svg")