Correct EASE R and update full results. Minor typos and formatting

Base/BaseMatrixFactorizationRecommender.py

@@ -88,7 +88,6 @@ def save_model(self, folder_path, file_name = None):
         data_dict_to_save = {"USER_factors": self.USER_factors,
                               "ITEM_factors": self.ITEM_factors,
                               "use_bias": self.use_bias,
-                              "_cold_user_mask": self._cold_user_mask,
                             }
 
         if self.use_bias:

Base/DataIO.py

@@ -119,7 +119,7 @@ def save_data(self, file_name, data_dict_to_save):
             current_file_path = current_temp_folder + attrib_name
 
             if isinstance(attrib_data, DataFrame):
-                attrib_data.to_csv(current_file_path, index=False)
+                attrib_data.to_csv(current_file_path + ".csv", index=False)
                 attribute_to_file_name[attrib_name] = attrib_name + ".csv"
 
             elif isinstance(attrib_data, sps.spmatrix):
@@ -132,8 +132,22 @@ def save_data(self, file_name, data_dict_to_save):
                 attribute_to_file_name[attrib_name] = attrib_name + ".npy"
 
             else:
-                attribute_to_json_file[attrib_name] = attrib_data
-                attribute_to_file_name[attrib_name] = attrib_name + ".json"
+                # Try to parse it as json, if it fails and the data is a dictionary, use another zip file
+                try:
+                    _ = json.dumps(attrib_data, default=json_not_serializable_handler)
+                    attribute_to_json_file[attrib_name] = attrib_data
+                    attribute_to_file_name[attrib_name] = attrib_name + ".json"
+
+                except TypeError:
+
+                    if isinstance(attrib_data, dict):
+                        dataIO = DataIO(folder_path = current_temp_folder)
+                        dataIO.save_data(file_name = attrib_name, data_dict_to_save=attrib_data)
+                        attribute_to_file_name[attrib_name] = attrib_name + ".zip"
+
+                    else:
+                        raise TypeError("Type not recognized for attribute: {}".format(attrib_name))
+
 
 
         # Save list objects
@@ -211,6 +225,10 @@ def load_data(self, file_name):
                     # allow_pickle is FALSE to prevent using pickle and ensure portability
                     attrib_data = np.load(attrib_file_path, allow_pickle=False)
 
+                elif attrib_data_type == "zip":
+                    dataIO = DataIO(folder_path = current_temp_folder)
+                    attrib_data = dataIO.load_data(file_name = file_name)
+
                 elif attrib_data_type == "json":
                     with open(attrib_file_path, "r") as json_file:
                         attrib_data = json.load(json_file)

Base/Evaluation/Evaluator.py

@@ -195,6 +195,10 @@ def __init__(self, URM_test_list, cutoff_list, min_ratings_per_user=1, exclude_s
 
             users_to_evaluate_mask = np.logical_or(users_to_evaluate_mask, new_mask)
 
+        if not np.all(users_to_evaluate_mask):
+            self._print("Ignoring {} ({:.2f}%) Users that have less than {} test interactions".format(np.sum(users_to_evaluate_mask),
+                                                                                                     100*np.sum(np.logical_not(users_to_evaluate_mask))/len(users_to_evaluate_mask), min_ratings_per_user))
+
         self.users_to_evaluate = np.arange(self.n_users)[users_to_evaluate_mask]
 
         if ignore_users is not None:

run_IJCAI_17_DELF.py

@@ -8,6 +8,7 @@
 
 import numpy as np
 import os, traceback, argparse
+import scipy.sparse as sps
 from functools import partial
 from Utils.plot_popularity import plot_popularity_bias, save_popularity_statistics
 
@@ -54,7 +55,11 @@ def cold_items_statistics(URM_train, URM_validation, URM_test, URM_test_negative
 
 
 
+def get_cold_items(URM):
 
+    cold_items_flag = np.ediff1d(sps.csc_matrix(URM).indptr) == 0
+
+    return np.arange(0, URM.shape[1])[cold_items_flag]
 
 
 def read_data_split_and_search(dataset_name,
@@ -68,14 +73,21 @@ def read_data_split_and_search(dataset_name,
         os.makedirs(result_folder_path)
 
 
+    # Ensure both experiments use the same data
+    dataset_folder_path = "result_experiments/{}/{}_{}/".format(CONFERENCE_NAME, ALGORITHM_NAME,
+                                                                dataset_name.replace("_remove_cold_items", ""))
+
+    if not os.path.exists(dataset_folder_path):
+        os.makedirs(dataset_folder_path)
+
     if 'amazon_music' in dataset_name:
-        dataset = AmazonMusicReader(result_folder_path)
+        dataset = AmazonMusicReader(dataset_folder_path)
 
-    elif 'movielens1m' in dataset_name:
-        dataset = Movielens1MReader(result_folder_path, type ="ours")
+    elif 'movielens1m_ours' in dataset_name:
+        dataset = Movielens1MReader(dataset_folder_path, type ="ours")
 
     elif 'movielens1m_original' in dataset_name:
-        dataset = Movielens1MReader(result_folder_path, type ="original")
+        dataset = Movielens1MReader(dataset_folder_path, type ="original")
 
     else:
         print("Dataset name not supported, current is {}".format(dataset_name))
@@ -101,11 +113,11 @@ def read_data_split_and_search(dataset_name,
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["Train data", "Test data"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
     save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
-                               ["Full data", "Train data", "Test data"],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")
 
 
@@ -135,9 +147,21 @@ def read_data_split_and_search(dataset_name,
     cutoff_list_validation = [10]
     cutoff_list_test = [5, 10, 20]
 
-    evaluator_validation = EvaluatorNegativeItemSample(URM_validation, URM_test_negative, cutoff_list=cutoff_list_validation)
-    evaluator_test = EvaluatorNegativeItemSample(URM_test, URM_test_negative, cutoff_list=cutoff_list_test)
+    if "_remove_cold_items" in dataset_name:
+        ignore_items_validation = get_cold_items(URM_train)
+        ignore_items_test = get_cold_items(URM_train + URM_validation)
+    else:
+        ignore_items_validation = None
+        ignore_items_test = None
+
+    evaluator_validation = EvaluatorNegativeItemSample(URM_validation, URM_test_negative, cutoff_list=cutoff_list_validation, ignore_items=ignore_items_validation)
+    evaluator_test = EvaluatorNegativeItemSample(URM_test, URM_test_negative, cutoff_list=cutoff_list_test, ignore_items=ignore_items_test)
 
+    # The Evaluator automatically skips users with no test interactions
+    # in this case we need the evaluation done with and without cold items to be comparable
+    # So we ensure the users that are included in the evaluation are the same in both cases.
+    evaluator_validation.users_to_evaluate = np.arange(URM_train.shape[0])
+    evaluator_test.users_to_evaluate = np.arange(URM_train.shape[0])
 
     runParameterSearch_Collaborative_partial = partial(runParameterSearch_Collaborative,
                                                        URM_train = URM_train,
@@ -289,7 +313,7 @@ def read_data_split_and_search(dataset_name,
 
     KNN_similarity_to_report_list = ["cosine", "dice", "jaccard", "asymmetric", "tversky"]
 
-    dataset_list = ['amazon_music', 'movielens1m']
+    dataset_list = ['amazon_music', 'movielens1m_ours', 'amazon_music_remove_cold_items', 'movielens1m_ours_remove_cold_items']
 
     for dataset_name in dataset_list:
         read_data_split_and_search(dataset_name,

run_IJCAI_17_DMF.py

@@ -109,11 +109,11 @@ def read_data_split_and_search(dataset_name,
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["URM train", "URM test"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
-    save_popularity_statistics([URM_train + URM_validation, URM_test],
-                               ["URM train", "URM test"],
+    save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")
 
 

run_IJCAI_18_NeuRec.py

@@ -415,7 +415,7 @@ def read_data_split_and_search(dataset_name,
 
         result_loader.generate_latex_results(file_name + "{}_latex_results.txt".format("beyond_accuracy_metrics"),
                                            metrics_list = ["DIVERSITY_MEAN_INTER_LIST", "DIVERSITY_HERFINDAHL", "COVERAGE_ITEM", "DIVERSITY_GINI", "SHANNON_ENTROPY"],
-                                           cutoffs_list = [50],
+                                           cutoffs_list = [10],
                                            table_title = None,
                                            highlight_best = True)
 

run_KDD_18_MCRec.py

@@ -57,11 +57,11 @@ def read_data_split_and_search(dataset_name,
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["URM train", "URM test"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
-    save_popularity_statistics([URM_train + URM_validation, URM_test],
-                               ["URM train", "URM test"],
+    save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")
 
 

run_RecSys_18_SpectralCF.py

@@ -140,11 +140,11 @@ def read_data_split_and_search(dataset_name, cold_start = False, cold_items=None
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["Train data", "Test data"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
     save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
-                               ["URM_all", "URM train", "URM test"],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")
 
 

run_SIGIR_18_CMN.py

@@ -102,11 +102,11 @@ def read_data_split_and_search(dataset_name,
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["URM train", "URM test"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
-    save_popularity_statistics([URM_train + URM_validation, URM_test],
-                               ["URM train", "URM test"],
+    save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")
 
 

run_WWW_17_NeuMF.py

@@ -69,11 +69,11 @@ def read_data_split_and_search(dataset_name,
     algorithm_dataset_string = "{}_{}_".format(ALGORITHM_NAME, dataset_name)
 
     plot_popularity_bias([URM_train + URM_validation, URM_test],
-                         ["URM train", "URM test"],
+                         ["Training data", "Test data"],
                          result_folder_path + algorithm_dataset_string + "popularity_plot")
 
-    save_popularity_statistics([URM_train + URM_validation, URM_test],
-                               ["URM train", "URM test"],
+    save_popularity_statistics([URM_train + URM_validation + URM_test, URM_train + URM_validation, URM_test],
+                               ["Full data", "Training data", "Test data"],
                                result_folder_path + algorithm_dataset_string + "popularity_statistics")