def get_clusters():
"""
Pimp cluster defs
"""
from pymeg import atlas_glasser as ag
# fmt: off
areas = ["47s", "47m", "a47r", "11l", "13l", "a10p", "p10p", "10pp", "10d", "OFC",
"pOFC", "44", "45", "IFJp", "IFJa", "IFSp", "IFSa", "47l", "p47r", "8C", "8Av",
"i6-8", "s6-8", "SFL", "8BL", "9p", "9a", "8Ad", "p9-46v", "a9-46v", "46",
"9-46d", "SCEF", "p32pr", "a24pr", "a32pr", "p24", "p32", "s32", "a24", "10v",
"10r", "25", "d32", "8BM", "9m"]
# fmt: on
areas = {
"NSWFRONT_" + area: ["L_{}_ROI-lh".format(area), "L_{}_ROI-lh".format(area)]
for area in areas
}
all_clusters, _, _, _ = ag.get_clusters()
all_clusters.update(areas)
all_clusters = {
k: v
for k, v in all_clusters.items()
if (k.startswith("NSWFRONT")) or (k in ag.areas.values())
}
return all_clusters
def aggregate(tfr_data, hemis, all_clusters=None):
"""Aggregate individual areas into clusters.
"""
from itertools import product
from pymeg import atlas_glasser
if all_clusters is None:
all_clusters, _, _, _ = atlas_glasser.get_clusters()
clusters = []
tfr_areas = np.unique(tfr_data.index.get_level_values('area'))
for hemi, cluster in product(hemis, all_clusters.keys()):
print('Working on %s, %s' % (hemi, cluster))
tfrs_rh = [area for area in all_clusters[cluster] if 'rh' in area]
tfrs_lh = [area for area in all_clusters[cluster] if 'lh' in area]
tfrs_rh = [
t for t in tfr_areas
if any([a.lower() in t.lower() for a in tfrs_rh])
]
tfrs_lh = [
t for t in tfr_areas
if any([a.lower() in t.lower() for a in tfrs_lh])
]
lh_idx = tfr_data.index.isin(tfrs_lh, level='area')
rh_idx = tfr_data.index.isin(tfrs_rh, level='area')
left = tfr_data.loc[lh_idx, :].groupby(['freq', 'trial']).mean()
right = tfr_data.loc[rh_idx, :].groupby(['freq', 'trial']).mean()
if hemi == 'Pair':
left.loc[:, 'cluster'] = cluster + '_LH'
left.loc[:, 'hemi'] = 'Pair'
right.loc[:, 'cluster'] = cluster + '_RH'
right.loc[:, 'hemi'] = 'Pair'
clusters.append(left)
clusters.append(right)
elif hemi == 'Single':
if len(left) > 0:
left.loc[:, 'cluster'] = cluster + '_LH'
left.loc[:, 'hemi'] = 'Single'
clusters.append(left)
elif len(right) > 0:
right.loc[:, 'cluster'] = cluster + '_RH'
right.loc[:, 'hemi'] = 'Single'
clusters.append(right)
else:
print('RH skipping', hemi, cluster, ' -> Seems to be empty')
print(left, right, lh_idx, rh_idx, tfrs_lh, tfrs_rh)
continue
else:
if hemi == 'Lateralized':
tfrs = left - right
elif hemi == 'Averaged':
tfrs = (right + left) / 2
tfrs.loc[:, 'cluster'] = cluster
tfrs.loc[:, 'hemi'] = hemi
clusters.append(tfrs)
df = pd.concat(clusters)
df.set_index(['cluster', 'hemi'], append=True, inplace=True)
return df.reorder_levels(['hemi', 'cluster', 'trial', 'freq'])
def plot_cluster(names, view):
from pymeg import atlas_glasser
all_clusters, _, _, _ = atlas_glasser.get_clusters()
label_names = []
for name in names:
cluster_name = atlas_glasser.areas[name]
label_names.extend(all_clusters[cluster_name])
plot_roi("lh", label_names, "r")
def aggregate(tfr_data, hemis, all_clusters=None):
"""Aggregate individual areas into clusters.
"""
from itertools import product
from pymeg import atlas_glasser
if all_clusters is None:
all_clusters, _, _, _ = atlas_glasser.get_clusters()
clusters = []
tfr_areas = np.unique(tfr_data.index.get_level_values("area"))
for hemi, cluster in product(hemis, all_clusters.keys()):
print("Working on %s, %s" % (hemi, cluster))
tfrs_rh = [area for area in all_clusters[cluster] if "rh" in area]
tfrs_lh = [area for area in all_clusters[cluster] if "lh" in area]
tfrs_rh = [
t for t in tfr_areas
if any([a.lower() in t.lower() for a in tfrs_rh])
]
tfrs_lh = [
t for t in tfr_areas
if any([a.lower() in t.lower() for a in tfrs_lh])
]
lh_idx = tfr_data.index.isin(tfrs_lh, level="area")
rh_idx = tfr_data.index.isin(tfrs_rh, level="area")
left = tfr_data.loc[lh_idx, :].groupby(["freq", "trial"]).mean()
right = tfr_data.loc[rh_idx, :].groupby(["freq", "trial"]).mean()
if (len(left) == 0) or (len(right) == 0):
print("%s: Left or right Hemi is empty, skipping for aggregation" %
(cluster))
continue
if hemi == "Pair":
left.loc[:, "cluster"] = cluster + "_LH"
left.loc[:, "hemi"] = "Pair"
right.loc[:, "cluster"] = cluster + "_RH"
right.loc[:, "hemi"] = "Pair"
clusters.append(left)
clusters.append(right)
else:
if hemi == "Lateralized":
tfrs = left - right
elif hemi == "Averaged":
tfrs = (right + left) / 2
tfrs.loc[:, "cluster"] = cluster
tfrs.loc[:, "hemi"] = hemi
clusters.append(tfrs)
df = pd.concat(clusters)
df.set_index(["cluster", "hemi"], append=True, inplace=True)
return df.reorder_levels(["hemi", "cluster", "trial", "freq"])
def compute_contrast(
contrasts,
data_globstring,
base_globstring,
meta_data,
baseline_time,
baseline_per_condition=True,
n_jobs=1,
cache=Cache(cache=False),
all_clusters=None,
):
"""Compute a single contrast from tfr data
Args:
contrast: dict
Contains contrast names as keys and len==3 tuples as values. The
tuples contain a list of condition names first, then a set of
weights for each condition, then the hemispheres to compute the
contrast across. Condition names identify columns in the meta
data that are one for each trial that belongs to this condition.
Hemispheres can be:
'lh_is_ipsi' if contrast is ipsi-contra hemi and left hemi is
ipsi.
'rh_is_ipsi' if contrast is ipis-contra and right hemi is ipsi
'avg' if contrast should be averaged across hemispheres
data_globstring: list
Each string in data_globstring selects a set of filenames if
passed through glob. Condition averages and baselines are then
computed for each group of filenames identified by one entry
in data_globstring. This is useful for, e.g. computing
conditions per session first, then averaging them and then
computing contrasts across sessions.
base_globstring: string or list
Same as data_globstring but selects data to use for baselining
meta_data: data frame
Meta data DataFrame with as many rows as trials.
baseline_time: tuple
all_clusters : dict with cluster definitions, default None
If None it is loaded from atlas_glasser get_contrasts
"""
from itertools import product
# load for all subjects:
tfr_condition = []
from functools import reduce
conditions = set(
reduce(lambda x, y: x + y, [x[0] for x in contrasts.values()]))
print("computing mean tfr for all areas and conditions...")
tfr_condition = pool_conditions(
conditions=conditions,
data_globs=data_globstring,
base_globs=base_globstring,
meta_data=meta_data,
baseline_time=baseline_time,
baseline_per_condition=baseline_per_condition,
n_jobs=n_jobs,
cache=cache,
)
print("computing contrasts for all clusters...")
# Lower case all area names
# FIXME: Set all area names to lower case!
if all_clusters is None:
all_clusters, _, _, _ = atlas_glasser.get_clusters()
tfr_areas = np.array([
a for a in tfr_condition.index.levels[np.where(
np.array(tfr_condition.index.names) == "area")[0][0]]
])
tfr_areas_lower = np.array([area.lower() for area in tfr_areas])
for cluster, areas in all_clusters.items():
new_areas = []
for area in areas:
idx = np.where(tfr_areas_lower == area.lower())[0]
if len(idx) == 1:
new_areas.append(tfr_areas[idx[0]])
all_clusters[cluster] = new_areas
# mean across sessions:
tfr_condition = tfr_condition.groupby(["area", "condition", "freq"]).mean()
cluster_contrasts = []
for cur_contrast in contrasts.items():
print(cur_contrast)
for cluster in all_clusters.keys():
print(cluster)
contrast, (conditions, weights, hemi) = cur_contrast
logging.info("Start computing contrast %s for cluster %s -> %s" %
(contrast, cluster, hemi))
right = []
left = []
for condition in conditions:
tfrs_rh = []
tfrs_lh = []
for area in all_clusters[cluster]:
area_idx = tfr_condition.index.isin([area], level="area")
condition_idx = tfr_condition.index.isin([condition],
level="condition")
subset = (tfr_condition.loc[area_idx
& condition_idx].groupby(
["freq"]).mean())
if "rh" in area:
tfrs_rh.append(subset)
else:
tfrs_lh.append(subset)
# What happens when an area is not defined for both hemis?
if (len(tfrs_lh) == 0) and (len(tfrs_rh) == 0):
logging.warn("Skipping condition %s in cluster %s" %
(condition, cluster))
continue
try:
left.append(pd.concat(tfrs_lh))
except ValueError:
print("Exception 327")
pass
try:
right.append(pd.concat(tfrs_rh))
except ValueError:
print("Exception 332")
pass
if (len(left) == 0) and (len(right) == 0):
logging.warn("Skipping cluster %s" % (cluster))
continue
if hemi == "rh_is_ipsi":
left, right = right, left
if "is_ipsi" in hemi:
if not len(left) == len(right):
logging.warn(
"Skipping cluster %s: does not have the same number of lh/rh rois"
% (cluster))
continue
tfrs = [left[i] - right[i] for i in range(len(left))]
else:
if (len(right) == 0) and (len(left) == len(weights)):
tfrs = left
elif (len(left) == 0) and (len(right) == len(weights)):
tfrs = right
else:
tfrs = [(right[i] + left[i]) / 2 for i in range(len(left))]
assert len(tfrs) == len(weights)
tfrs = [tfr * weight for tfr, weight in zip(tfrs, weights)]
tfrs = reduce(lambda x, y: x + y, tfrs)
tfrs = tfrs.groupby("freq").mean()
if tfrs.shape[0] == 0:
continue
tfrs.loc[:, "cluster"] = cluster
tfrs.loc[:, "contrast"] = contrast
tfrs.loc[:, "hemi"] = hemi
cluster_contrasts.append(tfrs)
logging.info("Done compute contrast")
return pd.concat(cluster_contrasts)
if __name__ == '__main__':
compute = True
plot = True
# subjects:
subjects = [
'jw01', 'jw02', 'jw03', 'jw05', 'jw07', 'jw08', 'jw09', 'jw10', 'jw11',
'jw12', 'jw13', 'jw14', 'jw15', 'jw16', 'jw17', 'jw18', 'jw19', 'jw20',
'jw21', 'jw22', 'jw23', 'jw24', 'jw30'
]
# subjects = ['jw01', 'jw02',]
# get clusters:
all_clusters, visual_field_clusters, glasser_clusters, jwg_clusters = atlas_glasser.get_clusters(
)
# define contrasts:
contrasts = {
'all': (['all'], [1]),
'choice': (['hit', 'fa', 'miss', 'cr'], (1, 1, -1, -1)),
'stimulus': (['hit', 'fa', 'miss', 'cr'], (1, -1, 1, -1)),
'hand': (['left', 'right'], (1, -1)),
'pupil': (['pupil_h', 'pupil_l'], (1, -1)),
}
# hemis:
hemis = ['avg', 'avg', 'avg', 'rh_is_ipsi', 'avg']
# compute contrasts:
if compute: