def save_popularity_statistics(URM_object_list, URM_name_list,
output_file_path):
shape = URM_object_list[0].shape
for URM_object in URM_object_list:
assert URM_object.shape == shape
n_items = shape[1]
ignore_items = np.array([])
latex_file = open(output_file_path, "w")
latex_file.write(
"\tMax pop \t&Min pop \t&Avg pop \t&Gini Index \t&Kendall Tau \t&Shannon \t&Herfindahl \\\\ \n"
)
base_item_popularity = None
for URM_index, URM_name in enumerate(URM_name_list):
URM_object = URM_object_list[URM_index]
URM_object = sps.csc_matrix(URM_object)
URM_string = "{}\t&".format(URM_name)
item_popularity = np.ediff1d(URM_object.indptr)
max_pop = item_popularity.max()
min_pop = item_popularity.min()
avg_pop = item_popularity.mean()
gini_index = gini(item_popularity)
if base_item_popularity is None:
base_item_popularity = item_popularity.copy()
kendall_tau, p_value = kendalltau(base_item_popularity,
item_popularity)
values_list = [max_pop, min_pop, avg_pop, gini_index, kendall_tau]
metric_object_dict = {
# "Gini Diversity": Gini_Diversity(n_items, ignore_items),
"Shannon Entropy": Shannon_Entropy(n_items, ignore_items),
"Herfindahl Index": Diversity_Herfindahl(n_items, ignore_items),
}
for metric_name, metric_object in metric_object_dict.items():
metric_object.recommended_counter = item_popularity.copy()
metric_value = metric_object.get_metric_value()
values_list.append(metric_value)
URM_string += "\t&".join("{:.2f}".format(el) for el in values_list)
latex_file.write(URM_string + "\\\\ \n")
def _create_empty_metrics_dict(cutoff_list, n_items, n_users, URM_train, URM_test, ignore_items, ignore_users, diversity_similarity_object):
empty_dict = {}
# global_RMSE_object = RMSE(URM_train + URM_test)
for cutoff in cutoff_list:
cutoff_dict = {}
for metric in EvaluatorMetrics:
if metric == EvaluatorMetrics.COVERAGE_ITEM:
cutoff_dict[metric.value] = Coverage_Item(n_items, ignore_items)
elif metric == EvaluatorMetrics.DIVERSITY_GINI:
cutoff_dict[metric.value] = Gini_Diversity(n_items, ignore_items)
elif metric == EvaluatorMetrics.SHANNON_ENTROPY:
cutoff_dict[metric.value] = Shannon_Entropy(n_items, ignore_items)
elif metric == EvaluatorMetrics.COVERAGE_USER:
cutoff_dict[metric.value] = Coverage_User(n_users, ignore_users)
elif metric == EvaluatorMetrics.DIVERSITY_MEAN_INTER_LIST:
cutoff_dict[metric.value] = Diversity_MeanInterList(n_items, cutoff)
elif metric == EvaluatorMetrics.DIVERSITY_HERFINDAHL:
cutoff_dict[metric.value] = Diversity_Herfindahl(n_items, ignore_items)
elif metric == EvaluatorMetrics.NOVELTY:
cutoff_dict[metric.value] = Novelty(URM_train)
elif metric == EvaluatorMetrics.AVERAGE_POPULARITY:
cutoff_dict[metric.value] = AveragePopularity(URM_train)
elif metric == EvaluatorMetrics.MAP:
cutoff_dict[metric.value] = MAP()
elif metric == EvaluatorMetrics.MRR:
cutoff_dict[metric.value] = MRR()
# elif metric == EvaluatorMetrics.RMSE:
# cutoff_dict[metric.value] = global_RMSE_object
elif metric == EvaluatorMetrics.DIVERSITY_SIMILARITY:
if diversity_similarity_object is not None:
cutoff_dict[metric.value] = copy.deepcopy(diversity_similarity_object)
else:
cutoff_dict[metric.value] = 0.0
empty_dict[cutoff] = cutoff_dict
return empty_dict
def test_Shannon_Entropy(self):
from Base.Evaluation.metrics import Shannon_Entropy
n_items = 1000
shannon_entropy = Shannon_Entropy(n_items, ignore_items=np.array([]))
shannon_entropy.recommended_counter = np.ones(n_items)
assert np.isclose(9.96, shannon_entropy.get_metric_value(), atol=1e-2), "metric incorrect"
shannon_entropy.recommended_counter = np.zeros(n_items)
shannon_entropy.recommended_counter[0] = 1.0
assert np.isclose(0.0, shannon_entropy.get_metric_value(), atol=1e-3), "metric incorrect"
shannon_entropy.recommended_counter = np.random.uniform(0, 100, n_items).astype(np.int)
assert np.isclose(9.6, shannon_entropy.get_metric_value(), atol=1e-1), "metric incorrect"
def create_empty_metrics_dict(n_items, n_users, URM_train, ignore_items,
ignore_users, cutoff,
diversity_similarity_object):
empty_dict = {}
# from Base.Evaluation.ResultMetric import ResultMetric
# empty_dict = ResultMetric()
for metric in EvaluatorMetrics:
if metric == EvaluatorMetrics.COVERAGE_ITEM:
empty_dict[metric.value] = Coverage_Item(n_items, ignore_items)
elif metric == EvaluatorMetrics.DIVERSITY_GINI:
empty_dict[metric.value] = Gini_Diversity(n_items, ignore_items)
elif metric == EvaluatorMetrics.SHANNON_ENTROPY:
empty_dict[metric.value] = Shannon_Entropy(n_items, ignore_items)
elif metric == EvaluatorMetrics.COVERAGE_USER:
empty_dict[metric.value] = Coverage_User(n_users, ignore_users)
elif metric == EvaluatorMetrics.DIVERSITY_MEAN_INTER_LIST:
empty_dict[metric.value] = Diversity_MeanInterList(n_items, cutoff)
elif metric == EvaluatorMetrics.DIVERSITY_HERFINDAHL:
empty_dict[metric.value] = Diversity_Herfindahl(
n_items, ignore_items)
elif metric == EvaluatorMetrics.NOVELTY:
empty_dict[metric.value] = Novelty(URM_train)
elif metric == EvaluatorMetrics.AVERAGE_POPULARITY:
empty_dict[metric.value] = AveragePopularity(URM_train)
elif metric == EvaluatorMetrics.MAP:
empty_dict[metric.value] = MAP()
elif metric == EvaluatorMetrics.MRR:
empty_dict[metric.value] = MRR()
elif metric == EvaluatorMetrics.DIVERSITY_SIMILARITY:
if diversity_similarity_object is not None:
empty_dict[metric.value] = copy.deepcopy(
diversity_similarity_object)
else:
empty_dict[metric.value] = 0.0
return empty_dict