def process(self):
"""
Evaluation function
:return: the overall value of Ranking-based Equal Opportunity (REO)
"""
for u, u_r in self._recommendations.items():
if len(self._relevance.get_user_rel(u)):
self.__user_pop_reo(u_r, set(self._train[u].keys()),
self._cutoff,
set(self._relevance.get_user_rel(u)))
PR = self._num / self._den
self._metric_objs_list = []
for i_category in range(self._item_n_clusters):
self._metric_objs_list.append(
ProxyMetric(
name=
f"REO-ProbToBeRanked_items:{self._item_clustering_name}-{i_category}",
val=PR[i_category],
needs_full_recommendations=False))
# Overall REO
self._metric_objs_list.append(
ProxyMetric(name=f"REO_items:{self._item_clustering_name}",
val=np.std(PR) / np.mean(PR),
needs_full_recommendations=False))
def process(self):
"""
Evaluation function
:return: the overall value of Bias Disparity - Bias Source
"""
for u, u_train in self._train.items():
self.__item_bias_disparity_bs(u, u_train)
clustering_count = Counter(self._item_clustering.values())
PC = np.array([
clustering_count.get(c, 0) /
len(self._item_clustering) if self._item_clustering else 1
for c in range(self._item_n_clusters)
])
self._BS = ((self._category_sum.T / self._total_sum).T) / PC
self._metric_objs_list = []
for u_group in range(self._user_n_clusters):
for i_category in range(self._item_n_clusters):
self._metric_objs_list.append(
ProxyMetric(
name=
f"BiasDisparityBS_users:{self._user_clustering_name}-{u_group}_items:{self._item_clustering_name}-{i_category}",
val=self._BS[u_group, i_category],
needs_full_recommendations=False))
def process(self):
"""
Evaluation function
:return: the overall value of Ranking-based Statistical Parity (RSP)
"""
for u, u_r in self._recommendations.items():
self.__user_pop_rsp(u_r, set(self._train[u].keys()), self._cutoff)
PR = self._num / self._den
self._metric_objs_list = []
for i_category in range(self._item_n_clusters):
self._metric_objs_list.append(
ProxyMetric(
name=
f"RSP-ProbToBeRanked_items:{self._item_clustering_name}-{i_category}",
val=PR[i_category],
needs_full_recommendations=False))
# Overall RSP
self._metric_objs_list.append(
ProxyMetric(name=f"RSP_items:{self._item_clustering_name}",
val=np.std(PR) / np.mean(PR),
needs_full_recommendations=False))
def process(self):
"""
Evaluation function
:return: the overall value of Bias Disparity
"""
BR = BiasDisparityBR(self._recommendations, self._config, self._params,
self._evaluation_objects,
self._additional_data).get_BR()
BS = BiasDisparityBS(self._recommendations, self._config, self._params,
self._evaluation_objects,
self._additional_data).get_BS()
BD = (BR - BS) / BS
self._metric_objs_list = []
for u_group in range(self._user_n_clusters):
for i_category in range(self._item_n_clusters):
self._metric_objs_list.append(
ProxyMetric(
name=
f"BiasDisparityBD_users:{self._user_clustering_name}-{u_group}_items:{self._item_clustering_name}-{i_category}",
val=BD[u_group, i_category],
needs_full_recommendations=False))