def extract_feature_dependent_feature(self,
extractor,
force_extraction=False,
verbose=0,
add_args=None,
custom_name=None):
"""
Extracts a feature which may be dependent on other features and stores it in the database
Parameters
----------
extractor : function, which takes the path of a data point, a dictionary of all other features and *args as
parameters and returns a feature
force_extraction : boolean, if True - will re-extract feature even if a feature with this name already
exists in the database, otherwise, will only extract if the feature doesn't exist in the database.
default value: False
verbose : int, if bigger than 0, will print the current number of the file for which data is being extracted
add_args : optional arguments for the extractor (list/dictionary/tuple/whatever). if None, the
extractor should take only one input argument - the file path. default value: None
custom_name : string, optional name for the feature (it will be stored in the database with the custom_name
instead of extractor function name). if None, the extractor function name will be used. default value: None
Returns
-------
None
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
return extract_feature_dependent_feature_base(
self.dbpath, self.path_to_set, self._set_object, extractor,
force_extraction, verbose, add_args, custom_name)
def return_labels_numpy(self, original=False):
"""
Returns a 2d numpy array of labels
Parameters
----------
original : if True, will return original labels, if False, will return transformed labels (as defined by
label_dict), default value: False
Returns
-------
A numpy array of labels, each row corresponds to a single datapoint
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
engine = create_engine('sqlite:////' + self.dbpath)
trainset.Base.metadata.create_all(engine)
session_cl = sessionmaker(bind=engine)
session = session_cl()
tmp_object = session.query(trainset.TrainSet).get(1)
columns_amt = len(tmp_object.labels['original'])
return_array = np.zeros([self.points_amt, columns_amt])
for i in enumerate(
session.query(trainset.TrainSet).order_by(
trainset.TrainSet.id)):
if original is False:
return_array[i[0], :] = i[1].labels['transformed']
else:
return_array[i[0], :] = i[1].labels['original']
session.close()
return return_array
def return_labels(self, original=False):
"""
Returns the labels of the dataset
Parameters
----------
original : if True, will return original labels, if False, will return transformed labels (as defined by
label_dict), default value: False
Returns
-------
A list of lists, each 'inside list' corresponds to a single data point, each element of the 'inside list' is a
label
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
engine = create_engine('sqlite:////' + self.dbpath)
trainset.Base.metadata.create_all(engine)
session_cl = sessionmaker(bind=engine)
session = session_cl()
return_list = []
for i in session.query(trainset.TrainSet).order_by(
trainset.TrainSet.id):
if original is True:
row_list = i.labels['original']
else:
row_list = i.labels['transformed']
return_list.append(row_list[:])
session.close()
return return_list
def return_real_id(self):
"""
Returns a list of real_id's
Parameters
----------
Returns
-------
A list of real_id values for the dataset (a real_id is the filename minus the suffix and prefix)
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
return return_real_id_base(self.dbpath, self._set_object)
def return_features(self, names='all'):
"""
Returns a list of extracted features from the database
Parameters
----------
names : list of strings, a list of feature names which are to be retrieved from the database, if equal
to 'all', the all features will be returned, default value: 'all'
Returns
-------
A list of lists, each 'inside list' corresponds to a single data point, each element of the 'inside list' is a
feature (can be of any type)
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
return return_features_base(self.dbpath, self._set_object, names)
def return_features_numpy(self, names='all'):
"""
Returns a 2d numpy array of extracted features
Parameters
----------
names : list of strings, a list of feature names which are to be retrieved from the database, if equal to 'all',
all features will be returned, default value: 'all'
Returns
-------
A numpy array of features, each row corresponds to a single datapoint. If a single feature is a 1d numpy array,
then it will be unrolled into the resulting array. Higher-dimensional numpy arrays are not supported.
"""
if self._prepopulated is False:
raise errors.EmptyDatabase(self.dbpath)
else:
return return_features_numpy_base(self.dbpath, self._set_object,
self.points_amt, names)