keys, values = options.keys(), options.values()
opts = [dict(zip(keys,[it for it in items])) for items in product(*values)]
# Permutation dataframe
p_values = []
for item in opts:
cond_dict = {k: v for k, v in item.items()}
item = {k: [v] for k, v in item.items()}
df_ = dataframe.copy()
data_ = table.copy()
data_ = filter_dataframe(data_, item)
item.update({'permutation':[0]})
df_ = filter_dataframe(df_, item)
df_avg = np.mean(df_['accuracy'].values)
p = (np.count_nonzero(data_['accuracy'].values > df_avg) + 1) / 100.
cond_dict['accuracy_perm'] = np.mean(data_['accuracy'].values)
cond_dict['accuracy_true'] = np.mean(df_['accuracy'].values)
cond_dict['p_value'] = p
p_values.append(cond_dict)
p_df = pd.DataFrame(p_values)
subjects = np.unique(event_file['name'].values)
path_wk = "/home/robbis/mount/meg_workstation/Carlo_MDM/%s/RESIDUALS_MVPA/beta_attributes_full.txt"
folders = os.listdir("/home/robbis/mount/meg_workstation/Carlo_MDM/")
name_dict = {}
for subj in subjects:
subdir = folders[np.nonzero([s.find(subj) != -1 for s in folders])[0][0]]
name_dict[subj] = subdir
df = filter_dataframe(event_file, name=[subj])
df.to_csv(path_wk % (subdir), index=False, sep=" ")
name_dict = {}
for subj in subjects:
subdir = folders[np.nonzero([s.find(subj) != -1 for s in folders])[0][0]]
name_dict[subj] = subdir
event_file['name'] = [name_dict[name] for name in event_file['name'].values]
from pyitab.analysis.results import get_results, df_fx_over_keys, \
get_permutation_values, filter_dataframe
import statsmodels.api as sm
dataframe = get_results(
'/media/robbis/DATA/fmri/carlo_ofp/0_results/review/',
dir_id="across",
field_list=['sample_slicer'],
)
df_lateral_ips = filter_dataframe(dataframe,
permutation=[0],
roi=['lateral_ips'])
key_across = ['roi_value', 'evidence', 'fold']
key_within = ['roi_value', 'evidence', 'subject']
df_lips_avg = df_fx_over_keys(df_lateral_ips,
keys=key_across,
attr=['score_accuracy'],
fx=np.mean)
key_value = 'accuracy'
results = []
dataframe_test = df_lips_avg
#dataframe_test = df_old
for roi in np.unique(dataframe_test['roi_value'].values):
mat = h5py.File(fname)
t = mat['timevec'][:].T[0]
idx = [0, 50, 100, 150, 200]
idx = [0, 20, 40, 60, 80, 100]
tt = t >= 0
xticklabels = np.array(["{:5.1f}".format(i) for i in t])
conditions = np.unique(average_df['targets'])
sessions = np.unique(average_df['ses'])
limits = (0.4, .95)
occurences = list(itertools.product(conditions, sessions))
for occurence in occurences:
df = filter_dataframe(average_df, targets=[occurence[0]], ses=[occurence[1]])
fig, axes = pl.subplots(2, 4, sharex=True)
for r, (_, row) in enumerate(df.iterrows()):
im = axes[0, r].imshow(row['score_score'],
origin='lower',
cmap=pl.cm.magma,
vmin=limits[0],
vmax=limits[1]
)
clf = row['estimator__clf']
kernel = ""
if r>2:
kernel = clf[clf.find("kernel='"):clf.find(", p")]
clf = clf[:clf.find("(")]+" "+kernel
axes[0, r].set_title("%s | %s" % (clf, occurence[0].replace("+", "-")))
mapping_type = {
'source-subject': 'seed',
'source-aal':'aal',
'sensor-subject': 'sensor'
}
dataframe['type'] = [mapping_type["%s-%s" % (d[1]['space'], d[1]['atlas'])] \
for d in dataframe.iterrows()]
fx_mean = lambda x : np.nanmean(np.dstack(x), axis=2).flatten()
i = 0
for t in np.unique(dataframe['type']):
for y in np.unique(dataframe['target_transformer__attr']):
df = filter_dataframe(dataframe, **{'type': [t],
'target_transformer__attr': [y],
'clf': ['Lasso', '6-Layer-NN', 'NL-SVR-10', 'L-SVR-10']
})
for score in ['r2']:
average_df = apply_function(df,
['sub', 'band', 'k', 'clf',
'target_transformer__attr'],
attr='score_%s' % (score),
fx=fx_mean)
grid = sns.FacetGrid(average_df,
col="band",
row="clf",
hue="sub",
palette="tab20c",