def add_metafeatures(pairs, data):
if not CONFIG.ADD_METAFEATURES:
return data
mf = metafeatures(pairs)
if CONFIG.COMPUTE_METAFEATURE_COMBINATIONS:
return to_2d(data, mf, column_combinations(data, mf))
else:
return to_2d(data, mf)
def add_metafeatures(pairs, data):
if not CONFIG.ADD_METAFEATURES:
return data
mf = metafeatures(pairs)
if CONFIG.COMPUTE_METAFEATURE_COMBINATIONS:
return to_2d(data, mf, column_combinations(data, mf))
else:
return to_2d(data, mf)
def aggregate_proxy(func, data, aggregate, **kwargs):
"""
performs the logic on whether or not to aggregate based on a boolean flag (aggregate)
"""
if aggregate:
# have each element be a column
return aggregate_apply(func, to_2d(data).T, **kwargs)
else:
assert len(data.shape) == 1
return func(data, **kwargs)
def aggregate_proxy(func, data, aggregate, **kwargs):
"""
performs the logic on whether or not to aggregate based on a boolean flag (aggregate)
"""
if aggregate:
# have each element be a column
return aggregate_apply(func, to_2d(data).T, **kwargs)
else:
assert len(data.shape) == 1
return func(data, **kwargs)
def aggregate_apply(func, items, **kwargs):
"""
performs a function with each item in a collection of items as a first argument,
then aggregating the results into a 1-D list
"""
preaggregate = []
for item in items:
preaggregate.append(func(item, **kwargs))
features = []
if len(preaggregate) > 0:
preaggregate2d = to_2d(np.array(preaggregate))
for feat in preaggregate2d.T:
for aggregator in CONFIG.AGGREGATORS:
result_append(features, aggregator(feat))
return features
def aggregate_apply(func, items, **kwargs):
"""
performs a function with each item in a collection of items as a first argument,
then aggregating the results into a 1-D list
"""
preaggregate = []
for item in items:
preaggregate.append(func(item, **kwargs))
features = []
if len(preaggregate) > 0:
preaggregate2d = to_2d(np.array(preaggregate))
for feat in preaggregate2d.T:
for aggregator in CONFIG.AGGREGATORS:
result_append(features, aggregator(feat))
return features
def estimator_features(A_feat, B_feat, current_type):
assert current_type in ("NN", "NC", "CN", "CC")
A_type, B_type = current_type
y = B_feat
if A_type == "N":
# convert numerical to 2-D matrix
X = to_2d(A_feat)
elif A_type == "C":
# convert categorical to binary matrix
X = LabelBinarizer().fit_transform(A_feat)
else:
raise Exception("improper A_type: {}".format(A_type))
if B_type == "N":
return regression_features(X, y)
elif B_type == "C":
return classification_features(X, y)
else:
raise Exception("improper B_type: {}".format(B_type))
def estimator_features(A_feat, B_feat, current_type):
assert current_type in ("NN", "NC", "CN", "CC")
A_type, B_type = current_type
y = B_feat
if A_type == "N":
# convert numerical to 2-D matrix
X = to_2d(A_feat)
elif A_type == "C":
# convert categorical to binary matrix
X = LabelBinarizer().fit_transform(A_feat)
else:
raise Exception("improper A_type: {}".format(A_type))
if B_type == "N":
return regression_features(X, y)
elif B_type == "C":
return classification_features(X, y)
else:
raise Exception("improper B_type: {}".format(B_type))
def estimator_features_wrapper(B_data, current_type):
assert len(B_data.shape) == 1
return estimator_features(to_2d(A_data), B_data, current_type)
def estimator_features_wrapper(B_data, current_type):
assert len(B_data.shape) == 1
return estimator_features(to_2d(A_data), B_data, current_type)