def combine_sessions(sessions, **kwargs):
""" Merge session data sets in single data set.
"""
# make a copy of the sessions, just to be safe
sessions_ = list(sessions)
# define dataset based on first session
dataset_ = None
# append data from other sessions
for i, session_ in enumerate(sessions_):
print("[+] session: {}, file: {}".format(
i, session_.meta.reset_index(drop=False).session[0])
)
if dataset_ is None:
dataset_ = Bunch(**dict(session_))
else:
dataset_.data = dataset_.data.append(session_.data, ignore_index=True, sort=False)
dataset_.meta = dataset_.meta.append(session_.meta, ignore_index=False, sort=False)
dataset_.tmask = dataset_.tmask.append(session_.tmask, ignore_index=False, sort=False)
# clean
if kwargs.get('clean_meta'):
dataset_.meta = clean_meta(dataset_.meta, **kwargs).reset_index(drop=False)
# set X, y
dataset_.X = dataset_.data.values.reshape(-1, dataset_.data.shape[-1])
dataset_.y = dataset_.meta.values.reshape(-1, dataset_.meta.shape[-1])
# cache sessions
dataset_.sessions = list(sessions)
return dataset_
def load_subject_meta(dataset, index=0, sessions=None, targets=None, **kwargs):
""" Load behavioral data for a single haxby subject. """
# load target information as string and give a numerical identifier to each
meta = pd.read_csv(dataset.session_target[index], sep=" ")
# condition mask
sessions = sessions or list(set(meta.chunks))
targets = targets or list(set(meta.labels))
# apply conditions mask
session_mask = meta.chunks.isin(sessions)
target_mask = meta.labels.isin(targets)
condition_mask = (session_mask & target_mask)
# mask, extract, factorize
target, session = meta.labels, meta.chunks
#target, target_names = pd.factorize(target)
target_names = np.ravel(TARGET_NAMES)
target = np.stack(list(map(TARGET_NAMES.index, target)))
meta = meta.assign(session=session, target=target)
# convert y to one-hot encoding
from sklearn.preprocessing import OneHotEncoder
encoder = OneHotEncoder(handle_unknown='ignore', sparse=False)
y = encoder.fit_transform(np.c_[target])
target = pd.DataFrame(y, columns=target_names)
# define colors, labels
target_colors = np.arange(len(target.columns))
# define colormap, norm
cmap = kwargs.get('cmap', 'nipy_spectral_r')
cmap = cmap if callable(cmap) else plt.get_cmap(cmap)
norm = mpl.colors.Normalize(target_colors.min(), target_colors.max() + 1)
# map colors, labels
cmap = mpl.colors.ListedColormap([cmap(norm(_)) for _ in target_colors])
target_colors_hex = [mpl.colors.to_hex(_) for _ in cmap.colors]
# for naming
session_code = "sess" + '_'.join(str(_) for _ in sessions)
# return as bunch
subject = Bunch()
subject.meta = meta.loc[condition_mask]
subject.target = target.loc[condition_mask]
subject.y = subject.meta.target
subject.groups = subject.meta.session
subject.condition_mask = condition_mask
subject.target_names = list(target_names)
subject.target_colors = list(target_colors_hex)
subject.cmap = cmap
subject.norm = norm
subject.session_code = session_code
return subject
