def main():
feature_dict, col_names = pp.set_targets('data/features.txt', threshold=-1)
# consider only the terms of interest
with open('data/terms.json', 'rb') as f:
terms = json.load(f)
for key in list(feature_dict):
feature_dict[key] = [x for x in terms if feature_dict[key][x] > THRESH]
if not feature_dict[key]:
del(feature_dict[key])
# filter coordinates based on voxels
coord_dict = ex.filter_studies_active_voxels('data/docdict.txt', 'data/MNI152_T1_2mm_brain.nii.gz',
threshold=500, radius=6)
# ensure that the keys are ints
for key in list(coord_dict):
if not isinstance(key, int):
coord_dict[int(key)] = coord_dict[key]
del(coord_dict[key])
# find intersecting dicts
coord_dict, feature_dict = ex.get_intersecting_dicts(coord_dict, feature_dict)
# get the respective vectors
X, y = pp.get_features_targets(coord_dict, feature_dict, labels=terms, mask='data/MNI152_T1_2mm_brain.nii.gz')
score_per_class, score_per_label = classify(X, y)
with open('class_scores.json', 'wb') as f:
json.dump(score_per_class, f)
with open('label_scores.json', 'wb') as f:
json.dump(score_per_label, f)
return
def filter_studies_terms(feature_file, terms=None, threshold=0.001,
set_unique_label=False):
"""
Given the frequency of terms corresponding to each study, as well as the
tems to consider, eliminates all studies that have more than one term
occuring at frequency >= threshold
Parameters
----------
feature_file : str
the file with the raw features.
terms : list of str, optional
the terms that are being considered as labels. If not specified,
uses the 25 terms from the original study.
threshold : real, optional
the frequency of the term for it to be considered, as significant with
respect to the study. If not specified, uses 0.001 as by the original
paper.
set_unique_label : bool, optional
defaults to false, when true returns only a single label corresponding
to each study
Returns
-------
feature_dict : dict
the dict such that all studies with conflicting labels are eliminated.
"""
if terms is None:
terms = ['Semantic',
'Encoding',
'Executive',
'Language',
'Verbal',
'Phonological',
'Visual',
'Inference',
'Working Memory',
'Conflict',
'Spatial',
'Attention',
'Imagery',
'Action',
'Sensory',
'Perception',
'Auditory',
'Pain',
'Reward',
'Arousal',
'Emotion',
'Social',
'Episodic',
'Retrieval',
'Recognition'
]
feature_dict, target_names = pp.set_targets(feature_file,
threshold=-1)
# validate that the terms are actual features and convert to lower case
new_terms = [x.lower() for x in terms if x.lower() in target_names]
for key in list(feature_dict.keys()):
# remove all studies that have more than one major term
if len([x for x in new_terms if feature_dict[key][x] >
threshold])>1:
del(feature_dict[key])
if set_unique_label:
for key in list(feature_dict):
vmax = 0
label = None
for x in new_terms:
if feature_dict[key][x] > vmax:
vmax = feature_dict[key][x]
label = x
if label is not None:
feature_dict[key] = label
else:
del(feature_dict[key])
return feature_dict
def filter_studies_terms(feature_file, terms=None, threshold=0.001,
set_unique_label=False):
"""
Given the frequency of terms corresponding to each study, as well as the
tems to consider, eliminates all studies that have more than one term
occuring at frequency >= threshold
Parameters
----------
feature_dict : str
the file with the raw features.
terms : list of str, optional
the terms that are being considered as labels. If not specified,
uses the 25 terms from the original study.
threshold : real, optional
the frequency of the term for it to be considered, as significant with
respect to the study. If not specified, uses 0.001 as by the original
paper.
set_unique_label : bool, optional
defaults to false, when true returns only a single label corresponding
to each study
Returns
-------
feature_dict : dict
the dict such that all studies with conflicting labels are eliminated.
"""
if terms is None:
terms = ['Semantic',
'Encoding',
'Executive',
'Language',
'Verbal',
'Phonological',
'Visual',
'Inference',
'Working Memory',
'Conflict',
'Spatial',
'Attention',
'Imagery',
'Action',
'Sensory',
'Perception',
'Auditory',
'Pain',
'Reward',
'Arousal',
'Emotion',
'Social',
'Episodic',
'Retrieval',
'Recognition'
]
feature_dict, target_names = pp.set_targets(feature_file,
threshold=-1)
# validate that the terms are actual features and convert to lower case
new_terms = [x.lower() for x in terms if x.lower() in target_names]
for key in list(feature_dict.keys()):
# remove all studies that have more than one major term
if len([x for x in new_terms if feature_dict[key][x] >
threshold])>1:
del(feature_dict[key])
if set_unique_label:
for key in list(feature_dict):
vmax = 0
label = None
for x in new_terms:
if feature_dict[key][x] > vmax:
vmax = feature_dict[key][x]
label = x
if label is not None:
feature_dict[key] = label
else:
del(feature_dict[key])
return feature_dict
