def md_with_dataframe():
cols = 'person_id question_code survey_code response Corps'.split()
recs = [
(3554131, 'CSI1', '1415MYS', 5, '1st year'),
(3490339, 'CSI1', '1415MYS', 3, '1st year'),
(3557769, 'CSI1', '1415MYS', 3, '1st year'),
(3013575, 'CSI1', '1415MYS', 5, '1st year'),
(3554131, 'CSI1', '1415EYS', 6, '1st year'),
(3490339, 'CSI1', '1415EYS', 6, '1st year'),
(3557769, 'CSI1', '1415EYS', 6, '1st year'),
(3013575, 'CSI1', '1415EYS', 5, '1st year'),
]
md = ModelData()
md.df = pd.DataFrame.from_records(recs, columns=cols)
return md
def fit(cache,
eid_set=None,
experiments_exclude=[],
high_res=False,
threshold_injection=True):
logging.debug('getting data')
ipsi_data = ModelData(cache, ROOT_ID).get_regional_data(
eid_set=eid_set,
experiments_exclude=experiments_exclude,
high_res=high_res,
threshold_injection=threshold_injection,
projection_hemisphere_id=2)
contra_data = ModelData(cache, ROOT_ID).get_regional_data(
eid_set=eid_set,
experiments_exclude=experiments_exclude,
high_res=high_res,
projection_hemisphere_id=1,
threshold_injection=threshold_injection)
X = ipsi_data.injections
y = np.hstack((ipsi_data.projections, contra_data.projections))
logging.debug('fitting')
reg = HomogeneousModel(kappa=np.inf)
reg.fit(X, y)
# get ids
ss_ids = get_summary_structure_ids(cache)
injection_key = ipsi_data.injection_mask.get_key(structure_ids=ss_ids,
hemisphere_id=2)
ipsi_key = ipsi_data.projection_mask.get_key(structure_ids=ss_ids,
hemisphere_id=2)
contra_key = contra_data.projection_mask.get_key(structure_ids=ss_ids,
hemisphere_id=1)
injection_regions = nonzero_unique(injection_key)
ipsi_regions = nonzero_unique(ipsi_key)
contra_regions = nonzero_unique(contra_key)
ipsi_w = pd.DataFrame(data=reg.weights[:, :len(ipsi_regions)],
index=injection_regions,
columns=ipsi_regions)
contra_w = pd.DataFrame(data=reg.weights[:, len(ipsi_regions):],
index=injection_regions,
columns=contra_regions)
return pd.concat((ipsi_w, contra_w), keys=('ipsi', 'contra'), axis=1)
def fit_structure(cache,
structure_id,
experiments_exclude,
kernel=None,
model_option='standard'):
data = ModelData(
cache,
structure_id).get_voxel_data(experiments_exclude=experiments_exclude)
# nested cross val
logging.debug("Performing cross validation: (%d samples, %d vars)",
*data.projections.shape)
error = VoxelModelError(cache, data, kernel=kernel)
reg = error.single_cv(option=model_option)
logging.debug("score : %.2f", reg.best_score_)
if kernel == 'polynomial':
results = dict(shape=reg.kernel.shape, support=reg.kernel.support)
logging.debug("optimal shape : %.0f", results['shape'])
logging.debug("optimal support: %.0f", results['support'])
else:
results = dict(gamma=reg.gamma)
logging.debug("optimal gamma : %.3f", results['gamma'])
logging.debug("(optimal sigma : %.3f)", 1 / sqrt(results['gamma']))
return results
def run_structure(cache, structure_id, kernel=None, cv=None, eid_set=None,
experiments_exclude=[], error_option='standard'):
data = ModelData(cache, structure_id).get_voxel_data(
eid_set=eid_set, experiments_exclude=experiments_exclude)
# nested cross val
logging.debug("Performing nested cross validation: (%d samples, %d vars)",
*data.projections.shape)
error = VoxelModelError(cache, data, cv=cv, kernel=kernel)
error.run(option=error_option)
return error.scores
def run_structure(cache, structure_id, cv=None, eid_set=None, experiments_exclude=[],
high_res=False, threshold_injection=True):
data = ModelData(cache, structure_id).get_regional_data(
eid_set=eid_set, experiments_exclude=experiments_exclude, high_res=high_res,
threshold_injection=threshold_injection)
# nested cross val
logging.debug("Performing nested cross validation: "
"(%s samples, %s vars)" % data.projections.shape)
error = HomogeneousModelError(cache, data, cv=cv)
error.run()
return error.scores
def fit_structure(cache,
structure_id,
experiments_exclude,
kernel_params,
model_option='standard'):
data = ModelData(
cache,
structure_id).get_voxel_data(experiments_exclude=experiments_exclude)
# nested cross val
nw_kwargs = dict()
if 'shape' in kernel_params:
nw_kwargs['kernel'] = Polynomial(**kernel_params)
else:
nw_kwargs['kernel'] = 'rbf'
nw_kwargs['gamma'] = kernel_params.pop('gamma')
error = VoxelModelError(cache, data)
return data, error.fit(**nw_kwargs, option=model_option)
def run_structure(cache,
structure_id,
eid_set=None,
experiments_exclude=[],
error_option='standard',
**nw_kwargs):
data = ModelData(cache, structure_id).get_voxel_data(
eid_set=eid_set, experiments_exclude=experiments_exclude)
# nested cross val
logging.debug("Performing cross validation: (%d samples, %d vars)",
*data.projections.shape)
scoring_dict = LogHybridScorer(cache).scoring_dict
reg = NadarayaWatson(**nw_kwargs)
scores = cross_validate(reg,
X=data.centroids,
y=data.projections,
cv=LeaveOneOut(),
scoring=scoring_dict,
return_train_score=True,
n_jobs=-1)
return scores