def _verify_data_consistency(self):
self._assert_is_initialized()
print_preamble = "{} consistency check: ".format(
self.DATA_SPLITTER_NAME)
URM_to_load_list = ["URM_train", "URM_test"]
if self.use_validation_set:
URM_to_load_list.append("URM_validation")
assert len(self.SPLIT_URM_DICT) == len(URM_to_load_list),\
print_preamble + "The available URM are not as many as they are supposed to be. URMs are {}, expected URMs are {}".format(len(self.SPLIT_URM_DICT), len(URM_to_load_list))
assert all(URM_name in self.SPLIT_URM_DICT
for URM_name in URM_to_load_list
), print_preamble + "Not all URMs have been created"
assert all(
URM_name in URM_to_load_list
for URM_name in self.SPLIT_URM_DICT.keys()
), print_preamble + "The split contains URMs that should not exist"
URM_shape = None
for URM_name, URM_object in self.SPLIT_URM_DICT.items():
if URM_shape is None:
URM_shape = URM_object.shape
n_users, n_items = URM_shape
assert n_users != 0, print_preamble + "Number of users in URM is 0"
assert n_items != 0, print_preamble + "Number of items in URM is 0"
assert URM_shape == URM_object.shape, print_preamble + "URM shape is inconsistent"
assert self.SPLIT_URM_DICT[
"URM_train"].nnz != 0, print_preamble + "Number of interactions in URM Train is 0"
assert self.SPLIT_URM_DICT[
"URM_test"].nnz != 0, print_preamble + "Number of interactions in URM Test is 0"
URM = self.SPLIT_URM_DICT["URM_test"].copy()
user_interactions = np.ediff1d(sps.csr_matrix(URM).indptr)
assert np.all(
user_interactions == self.k_out_value
), print_preamble + "Not all users have the desired number of interactions in URM_test, {} users out of {}".format(
(user_interactions != self.k_out_value).sum(), n_users)
if self.use_validation_set:
assert self.SPLIT_URM_DICT[
"URM_validation"].nnz != 0, print_preamble + "Number of interactions in URM Validation is 0"
URM = self.SPLIT_URM_DICT["URM_validation"].copy()
user_interactions = np.ediff1d(sps.csr_matrix(URM).indptr)
assert np.all(
user_interactions == self.k_out_value
), print_preamble + "Not all users have the desired number of interactions in URM_validation, {} users out of {}".format(
(user_interactions != self.k_out_value).sum(), n_users)
URM = self.SPLIT_URM_DICT["URM_train"].copy()
user_interactions = np.ediff1d(sps.csr_matrix(URM).indptr)
if not self.allow_cold_users:
assert np.all(
user_interactions != 0
), print_preamble + "Cold users exist despite not being allowed as per DataSplitter parameters, {} users out of {}".format(
(user_interactions == 0).sum(), n_users)
assert assert_disjoint_matrices(list(self.SPLIT_URM_DICT.values()))
assert_URM_ICM_mapper_consistency(
URM_DICT=self.SPLIT_URM_DICT,
GLOBAL_MAPPER_DICT=self.SPLIT_GLOBAL_MAPPER_DICT,
ICM_DICT=self.SPLIT_ICM_DICT,
ICM_MAPPER_DICT=self.SPLIT_ICM_MAPPER_DICT,
DATA_SPLITTER_NAME=self.DATA_SPLITTER_NAME)
def _verify_data_consistency(self):
self._assert_is_initialized()
print_preamble = "{} consistency check: ".format(
self.DATA_SPLITTER_NAME)
assert len(self.SPLIT_URM_DICT) == len(self._SPLIT_URM_NAME_LIST),\
print_preamble + "The available URM are not as many as they are supposed to be. URMs are {}, expected URMs are {}".format(len(self.SPLIT_URM_DICT), len(self._SPLIT_URM_NAME_LIST))
assert all(URM_name in self.SPLIT_URM_DICT
for URM_name in self._SPLIT_URM_NAME_LIST
), print_preamble + "Not all URMs have been created"
assert all(
URM_name in self._SPLIT_URM_NAME_LIST
for URM_name in self.SPLIT_URM_DICT.keys()
), print_preamble + "The split contains URMs that should not exist"
URM_shape = None
for URM_name, URM_object in self.SPLIT_URM_DICT.items():
if URM_shape is None:
URM_shape = URM_object.shape
n_users, n_items = URM_shape
assert n_users != 0, print_preamble + "Number of users in URM is 0"
assert n_items != 0, print_preamble + "Number of items in URM is 0"
assert URM_shape == URM_object.shape, print_preamble + "URM shape is inconsistent"
assert self.SPLIT_URM_DICT[
"URM_train"].nnz != 0, print_preamble + "Number of interactions in URM Train is 0"
assert self.SPLIT_URM_DICT[
"URM_test"].nnz != 0, print_preamble + "Number of interactions in URM Test is 0"
# Assert URM_validation is not empty only when the input quota list is zero
# It may create problems on the user-wise split if the validation quota is too small and no items gets selected
# Although we assume in that case it would be acceptable to receive a warning your validation data cannot be built
assert (self.SPLIT_URM_DICT["URM_validation"].nnz == 0 and self.input_split_interaction_quota_list[1] == 0.0)\
or (self.SPLIT_URM_DICT["URM_validation"].nnz != 0 and self.input_split_interaction_quota_list[1] > 0.0)\
, print_preamble + "Number of interactions in Validation is 0"
quota_oscillation_allowed = 0.2
for URM_index, URM_name in enumerate(self._SPLIT_URM_NAME_LIST):
input_quota = self.input_split_interaction_quota_list[URM_index]
actual_quota = self.actual_split_interaction_quota_list[URM_index]
max_value_allowed = input_quota * (1 + quota_oscillation_allowed)
min_value_allowed = input_quota * (1 - quota_oscillation_allowed)
if actual_quota < min_value_allowed or actual_quota > max_value_allowed:
print(
print_preamble +
"The differentce between the input interaction quota '{}' and actual interaction quota '{}' of '{}' higher than {} %"
.format(input_quota, actual_quota, URM_name,
quota_oscillation_allowed * 100))
URM = self.SPLIT_URM_DICT["URM_train"].copy()
user_interactions = np.ediff1d(sps.csr_matrix(URM).indptr)
if not self.allow_cold_users:
assert np.all(
user_interactions != 0
), print_preamble + "Cold users exist despite not being allowed as per DataSplitter parameters, {} users out of {}".format(
(user_interactions == 0).sum(), n_users)
assert assert_disjoint_matrices(list(self.SPLIT_URM_DICT.values()))
assert_URM_ICM_mapper_consistency(
URM_DICT=self.SPLIT_URM_DICT,
user_original_ID_to_index=self.
SPLIT_GLOBAL_MAPPER_DICT["user_original_ID_to_index"],
item_original_ID_to_index=self.
SPLIT_GLOBAL_MAPPER_DICT["item_original_ID_to_index"],
ICM_DICT=self.SPLIT_ICM_DICT,
ICM_MAPPER_DICT=self.SPLIT_ICM_MAPPER_DICT,
UCM_DICT=self.SPLIT_UCM_DICT,
UCM_MAPPER_DICT=self.SPLIT_UCM_MAPPER_DICT,
DATA_SPLITTER_NAME=self.DATA_SPLITTER_NAME)