def get_centers(n_cluster, **kwargs):
conf = {
'path': '/media/robbis/DATA/fmri/movie_viviana/',
'filetype': 'masked',
'fname': 'mat_corr_sub_%s.mat',
'band': 'alpha',
'conditions': ['movie', 'scramble', 'rest'],
'state_res_fname': "clustering_labels_%s_maxk_%s_%s_%s.pyobj",
'max_k': 15,
'method': 'speed'
}
conditions = conf['conditions']
conf.update(kwargs)
centers = []
_, clustering = get_results(**conf)
data_fname = os.path.join(conf['path'], conf['band'], conf['filetype'],
conf['fname'])
for c in conditions:
data_, _ = get_data(data_fname % str(c))
X = get_subsampler(conf['method']).fit_transform(data_)
centroids = get_centroids(X, clustering[c][n_cluster - 2])
centers.append(centroids)
return centers, conditions
def demeaned():
conf = {
'path': '/media/robbis/DATA/fmri/movie_viviana/',
'filetype': 'masked',
'fname': 'mat_corr_sub_%s.mat',
'conditions': ['movie', 'scramble', 'rest'],
'state_res_fname': "clustering_labels_%s_maxk_%s_%s_%s.pyobj",
'max_k': 15,
'method': 'speed'
}
filename = os.path.join(conf['path'], conf['filetype'], conf['fname'])
for condition in conf["conditions"]:
data, n_roi = get_data(filename % (str(condition)))
#data = remove_component(data, method="mean", argument="session")
data = data - data.mean(1)[:, None, :]
X, clustering_ = cluster_state(data, range(2, 8), method='variance')
metrics, k_step, metric_names = calculate_metrics(X, clustering_)
clustering_ = pickle.load(file(label_fname % (condition.lower()), 'r'))
centroid_ = get_centroids(X, clustering_[3]) # Five centroids
centroid_ = centroid_[order[condition], :] # Similarity reorder
dict_centroids[condition.lower()] = centroid_
get_positions(dict_centroids, path)
def demeaned():
conf = {'path':'/media/robbis/DATA/fmri/movie_viviana/',
'filetype' : 'masked',
'fname': 'mat_corr_sub_%s.mat',
'conditions' : ['movie', 'scramble', 'rest'],
'state_res_fname' : "clustering_labels_%s_maxk_%s_%s_%s.pyobj",
'max_k':15,
'method':'speed'
}
filename = os.path.join(conf['path'], conf['filetype'], conf['fname'])
for condition in conf["conditions"]:
data, n_roi = get_data(filename % (str(condition)))
#data = remove_component(data, method="mean", argument="session")
data = data - data.mean(1)[:,None,:]
X, clustering_ = cluster_state(data, range(2,8), method='variance')
metrics, k_step, metric_names = calculate_metrics(X, clustering_)
clustering_ = pickle.load(file(label_fname % (condition.lower()), 'r'))
centroid_ = get_centroids(X, clustering_[3]) # Five centroids
centroid_ = centroid_[order[condition],:] # Similarity reorder
dict_centroids[condition.lower()] = centroid_
get_positions(dict_centroids, path)
def analysis(**kwargs):
"""
Method to fast analyze states from matrices.
Parameters
----------
kwargs : dictionary of several parameters
path : string of the data path
filetype : string ('masked', 'original')
if you want to use masked matrix or full rank.
fname : string. pattern of the mat file in input
default ("mat_corr_sub_%s.mat")
conditions : list of string of data conditions.
should cope with file pattern specified in fname.
method : string ('speed', 'variance', 'variance+mean', 'speed+mean')
Method used to subsample data. Using '+mean' suffix let
to select only points that are greater than mean.
max_k : integer (default = 15).
The maximum number of cluster to use
filter : string ('none', 'normalize')
Method used to normalize and filter data.
state_res_fname : pattern of the output file
(default "clustering_labels_%s_maxk_%s_%s_%s.pyobj")
File used to save labels after clustering.
"""
from mvpa_itab.conn.states.plot import plot_metrics, plot_states_matrices
configuration = {
'path': '/media/robbis/DATA/fmri/movie_viviana/',
'band': 'alpha',
'filetype': 'masked',
'fname': 'mat_corr_sub_%s.mat',
'conditions': ['movie', 'scramble', 'rest'],
'state_res_fname': "clustering_labels_%s_maxk_%s_%s_%s.pyobj",
'max_k': 15,
'method': 'variance+mean',
'filter': 'none'
}
configuration.update(kwargs)
logger.info(configuration)
conditions = configuration['conditions']
max_k = configuration['max_k']
method = configuration['method']
filetype = configuration['filetype']
filter_type = configuration['filter']
data_filter = get_filter(configuration['filter'])
result = dict()
path = os.path.join(configuration['path'], configuration['band'],
configuration['filetype'])
for cond in conditions:
data_, n_roi = get_data(
os.path.join(path, "mat_corr_sub_%s.mat" % (str(cond))))
data_ = data_filter.transform(data_)
X = get_subsampler(method).fit_transform(data_)
X, clustering_ = cluster_state(X, range(2, max_k))
metrics_, k_step, metrics_keys = calculate_metrics(X, clustering_)
fig = plot_metrics(metrics_, metrics_keys, k_step)
directory = os.path.join(path, method, filter_type, cond)
if not os.path.exists(directory):
os.makedirs(directory)
# Save results
state_res_fname = configuration['state_res_fname']
pickle.dump(
clustering_,
file(
os.path.join(
directory,
state_res_fname % (cond, str(max_k), method, filetype),
), 'w'))
# Save metrics
fig_fname = os.path.join(directory, "metrics_k_%s.png" % (str(max_k)))
fig.savefig(fig_fname)
#save_data(X, 'metrics', metrics_)
for i, labels in enumerate(clustering_):
if not os.path.exists(os.path.join(directory, str(i + 2))):
os.makedirs(os.path.join(directory, str(i + 2)))
plot_states_matrices(X,
labels,
save_path=os.path.join(directory, str(i + 2)),
save_name_condition=cond)
# Save filtered data
X_fname = os.path.join(directory, "filtered_data.mat")
save_data(X, 'filtered_data', X_fname)
result[cond] = {'X': X, 'clustering': clustering_, 'data': data_}
return result
import itertools
from sklearn.manifold.mds import MDS
"""
distance_fname = os.path.join('/media/robbis/DATA/fmri/movie_viviana/',
'pairwise_distance_all_subj_all_pts_movie.npy')
clustering_ = pickle.load(file('/media/robbis/DATA/fmri/movie_viviana/clustering_labels_2to30k_movie_all_sub_all_pts.obj', 'r'))
distance_matrix = np.load(distance_fname, mmap_mode='r')
"""
path = "/media/robbis/DATA/fmri/movie_viviana/masked/"
conditions = ['rest', 'movie', 'scramble']
filetype = 'masked'
for cond in conditions:
data_, n_roi = get_data(os.path.join(path,"mat_corr_sub_%s_masked.mat" % (str.upper(cond))))
max_k = 15
method = 'speed'
X, clustering_ = cluster_state(data_, max_k, method)
metrics_, k_step, metrics_keys = calculate_metrics(X, clustering_)
pickle.dump(clustering_, file(os.path.join(path,
"clustering_labels_%s_maxk_%s_%s_%s.pyobj" %(cond,
str(max_k),
method,
filetype),
),
'w'))
method = ['speed', 'variance']
conditions = ['movie', 'scramble', 'rest']
X_centers = dict()
prod = itertools.product(filetypes, method, conditions)
for t, m, c in prod:
path_cluster = "%s/%s/%s" % (t, m, c)
path_cluster = os.path.join(configuration['path'], path_cluster)
fname = configuration['state_res_fname'] % (c, str(10), m, t)
path_file = os.path.join(path_cluster, fname)
clustering_ = pickle.load(file(path_file, 'r'))
data_fname = os.path.join(configuration['path'], t, configuration['fname'])
data_, _ = get_data(data_fname % str.upper(c))
subj_min_speed, subj_speed = get_subsampling_measure(m)(data_)
X = data_[subj_min_speed]
metrics_, k_step, metric_names = calculate_metrics(X, clustering_)
fig = plot_metrics(metrics_, metric_names, k_step)
fig_fname = os.path.join(configuration['path'], t, m,
m + "_metric_evaluation.png")
fig.savefig(fig_fname)
#####################
centers, conditions = get_centers(n_cluster=5,