Merge remote-tracking branch 'origin/sir_elliot'

.gitignore

@@ -20,8 +20,6 @@ pip-selfcheck.json
 
 /results/
 /data/*
-!/data/cat_dbpedia_movielens_1m
-!/data/movielens_1m
 /share/
 **/__pycache__/**
 *.log
@@ -47,3 +45,13 @@ MANIFEST
 docs/source/_build
 docs/source/_static
 docs/source/_templates
+
+
+!/data/cat_dbpedia_movielens_1m/features.tsv
+!/data/cat_dbpedia_movielens_1m/map.tsv
+!/data/cat_dbpedia_movielens_1m/properties.conf
+!/data/movielens_1m/i_pop.tsv
+!/data/movielens_1m/u_pop.tsv
+!/data/cat_dbpedia_movielens_1m_v030/features.tsv
+!/data/cat_dbpedia_movielens_1m_v030/map.tsv
+!/data/cat_dbpedia_movielens_1m_v030/properties.conf

CHANGELOG.md

@@ -2,6 +2,32 @@
 
 All notable changes to this project will be documented in this file
 
+## [v0.3.0] - 2020-06-30
+### Changed
+- early stopping strategies
+- offline recommendation files evaluation (ProxyRecommender, RecommendationFolder) 
+- negative sampling evaluation
+- improved Microsoft Windows compatibility  
+- binarization of explicit dataset
+- automatic loading of implicit datasets 
+- multiple prefiltering
+- managing side information with modular loaders
+- alignment of side information with training data
+- improved Documentation: Model creation, Side Information loading, Early Stopping, Negative Sampling 
+- added nDCG as formulated in Rendle's 2020 KDD paper
+- visual loader with tensorflow pipeline 
+- added and fixing visual recsys method:
+  - DVBPR
+  - VBPR
+  - DeepStyle
+  - ACF
+  - VNPR
+- added new recommender method
+  - MF (Rendle's 2020 RecSys reproducibility paper)
+  - EASER
+  - RP3beta
+  - iALS
+
 ## [v0.2.1] - 2020-03-27
 ### Changed
 

basic_configuration.md

@@ -17,6 +17,6 @@ by merely passing a list of possible hyperparameter values, e.g., neighbors: [50
 
 The reported models are selected according to nDCG@10.
 
-|To see the full configuration file please visit the following [link](config_files/basic_configuration.yml)|
+|To see the full configuration file please visit the following [link](config_files/basic_configuration_v030.yml)|
 |-------------------------------------------------------------------------------------------------------------|
 |**To run the experiment use the following [script](sample_basic.py)**|

config_files/advanced_configuration.yml

@@ -2,12 +2,11 @@ experiment:
   dataset: movielens_1m
   data_config:
     strategy: dataset
-    dataset_path: ../data/movielens_1m/dataset.tsv
+    dataset_path: ../data/{0}/dataset.tsv
   prefiltering:
     strategy: iterative_k_core
     core: 10
   splitting:
-    save_folder: ../data/movielens_1m/splitting/
     test_splitting:
         strategy: random_subsampling
         folds: 1
@@ -17,28 +16,14 @@ experiment:
         folds: 5
   top_k: 50
   evaluation:
-    simple_metrics: [nDCG, ACLT, APLT, ARP, PopREO]
+    simple_metrics: [nDCG, ACLT, ARP]
     complex_metrics:
     - metric: UserMADrating
       clustering_name: Happiness
       clustering_file: ../data/movielens_1m/u_happy.tsv
-    - metric: ItemMADrating
-      clustering_name: ItemPopularity
-      clustering_file: ../data/movielens_1m/i_pop.tsv
-    - metric: REO
-      clustering_name: ItemPopularity
-      clustering_file: ../data/movielens_1m/i_pop.tsv
-    - metric: RSP
-      clustering_name: ItemPopularity
-      clustering_file: ../data/movielens_1m/i_pop.tsv
-    - metric: BiasDisparityBD
-      user_clustering_name: Happiness
-      user_clustering_file: ../data/movielens_1m/u_happy.tsv
-      item_clustering_name: ItemPopularity
-      item_clustering_file: ../data/movielens_1m/i_pop.tsv
     relevance_threshold: 1
     wilcoxon_test: True
-  gpu: 1
+  gpu: 0
   models:
     NeuMF:
       meta:

config_files/basic_configuration.yml

@@ -1,24 +1,20 @@
 experiment:
+  version: 0.2.1
   dataset: cat_dbpedia_movielens_1m
   data_config:
     strategy: dataset
     dataloader: KnowledgeChainsLoader
-    dataset_path: ../data/cat_dbpedia_movielens_1m/dataset.tsv
+    dataset_path: ../data/{0}/dataset.tsv
     side_information:
         map: ../data/cat_dbpedia_movielens_1m/map.tsv
         features: ../data/cat_dbpedia_movielens_1m/features.tsv
         properties: ../data/cat_dbpedia_movielens_1m/properties.conf
 #  prefiltering:
 #    strategy: user_average # Not applied in the paper experiments
   splitting:
-    save_on_disk: True
-    save_folder: ../data/cat_dbpedia_movielens_1m/splitting/
     test_splitting:
         strategy: temporal_hold_out
         test_ratio: 0.2
-    validation_splitting:
-        strategy: temporal_hold_out
-        test_ratio: 0.2
   top_k: 50
   evaluation:
     cutoffs: [10, 5]

config_files/basic_configuration_v030.yml

@@ -0,0 +1,56 @@
+experiment:
+  version: 0.3.0
+  dataset: cat_dbpedia_movielens_1m
+  data_config:
+    strategy: dataset
+    dataset_path: ../data/cat_dbpedia_movielens_1m/dataset.tsv
+    side_information:
+      - dataloader: ChainedKG
+        map: ../data/cat_dbpedia_movielens_1m/map.tsv
+        features: ../data/cat_dbpedia_movielens_1m/features.tsv
+        properties: ../data/cat_dbpedia_movielens_1m/properties.conf
+#  prefiltering:
+#    strategy: user_average # Not applied in the paper experiments
+  splitting:
+    save_on_disk: True
+    save_folder: ../data/cat_dbpedia_movielens_1m/splitting/
+    test_splitting:
+        strategy: temporal_hold_out
+        test_ratio: 0.2
+    validation_splitting:
+        strategy: temporal_hold_out
+        test_ratio: 0.2
+  top_k: 50
+  evaluation:
+    cutoffs: [10, 5]
+    simple_metrics: [nDCG,Precision,ItemCoverage,EPC,Gini]
+    relevance_threshold: 1
+  gpu: 1
+  external_models_path: ../external/models/__init__.py
+  models:
+    Random:
+      meta:
+        verbose: True
+        save_recs: True
+      seed: 42
+    external.MostPop:
+      meta:
+        verbose: True
+        save_recs: True
+        validation_metric: nDCG@10
+    ItemKNN:
+      meta:
+        verbose: True
+        save_recs: True
+        validation_metric: nDCG@10
+      neighbors: [50, 70, 100]
+      similarity: [cosine, euclidean]
+      implementation: standard
+    AttributeItemKNN:
+      meta:
+        verbose: True
+        save_recs: True
+        validation_metric: nDCG@10
+      loader: ChainedKG
+      neighbors: [50, 70, 100]
+      similarity: [braycurtis, manhattan]

config_files/recsys_config.yml

@@ -0,0 +1,133 @@
+experiment:
+  dataset: ncf_ml1m
+  data_config:
+    strategy: fixed
+    train_path: ../data/{0}/ml-1m.train.rating
+    test_path: ../data/{0}/ml-1m.test.rating
+  binarize: True
+  negative_sampling:
+    strategy: fixed
+    files: [ "/home/ironman/PycharmProjects/Elliot/data/ncf_ml1m/ml-1m.test.negative" ]
+  top_k: 10
+  evaluation:
+    cutoffs: 10
+    simple_metrics: [nDCG, Recall, HR, Precision, MAP, MRR]
+  gpu: 0
+  external_models_path: ../external/models/__init__.py
+  models:
+#    Random:
+#      meta:
+#        save_recs: True
+#    external.MostPop:
+#      meta:
+#        verbose: True
+#        save_recs: True
+#    external.RendleMF: # from original paper
+#      meta:
+#        hyper_max_evals: 1
+#        hyper_opt_alg: tpe
+#        validation_rate: 1
+#        verbose: True
+#        save_recs: True
+        # optimize_internal_loss: True
+#      epochs: 256 # 256 original paper but 50 comes from NeuMF paper
+#      factors: 192
+#      lr: 0.002
+#      reg: 0.005
+#      m: 8
+#      random_seed: 42
+#    external.iALS: #from TOIS
+#      meta:
+#        hyper_max_evals: 20
+#        hyper_opt_alg: tpe
+#        verbose: True
+#        save_recs: True
+#        validation_rate: 20
+#      epochs: [uniform, 1, 500]
+#      scaling: [linear, log]
+#      factors: [uniform, 1, 200]
+#      alpha: [uniform, 10e-3, 50]
+#      epsilon: [uniform, 10e-3, 10]
+#      reg: [uniform, 10e-3, 10e-2]
+#    external.NeuMF: #from the original paper + Rendle
+#      meta:
+#        hyper_max_evals: 1
+#        hyper_opt_alg: tpe
+#        verbose: True
+#        save_recs: True
+#        validation_rate: 1
+#        optimize_internal_loss: True
+#      mf_factors:  64
+#      dropout: 0
+#      is_mf_train: True
+#      is_mlp_train: True
+#      batch_size: 256
+#      epochs: 100
+#      lr: 0.001
+#      m: 4
+#    ItemKNN: #from TOIS
+#      meta:
+#        save_recs: True
+#        verbose: True
+#        hyper_max_evals: 20
+#        hyper_opt_alg: tpe
+#      neighbors: [uniform, 5, 1000]
+#      similarity: [cosine, jaccard, dice, mahalanobis, euclidean]
+#    UserKNN: #from TOIS
+#      meta:
+#        hyper_max_evals: 20
+#        hyper_opt_alg: tpe
+#        save_recs: True
+#        verbose: True
+#      neighbors: [ uniform, 5, 1000 ]
+#      similarity: [cosine, jaccard, dice, mahalanobis, euclidean]
+#    MultiVAE: # from original paper
+#      meta:
+#        hyper_max_evals: 20
+#        hyper_opt_alg: tpe
+#        save_recs: True
+#        verbose: True
+#        # optimize_internal_loss: True
+#      lr: [loguniform, -11.512925464970229, 0] # exploration taken from TOIS
+#      epochs: 200
+#      batch_size: [ 128, 256, 512 ]
+#      intermediate_dim: 600
+#      latent_dim: 200
+#      dropout_pkeep: 0.5
+#      reg_lambda: [loguniform, -11.512925464970229, 0] # exploration taken from TOIS
+#    Slim: #from TOIS
+#      meta:
+#        hyper_max_evals: 1
+#        hyper_opt_alg: tpe
+#        verbose: True
+#        save_recs: True
+#      l1_ratio: 0.0000119
+#      alpha: 0.0788
+#      neighborhood: 544
+#    external.iALS:
+#      meta:
+#        verbose: True
+#        save_recs: True
+#      factors: 46
+#      alpha: 50
+#      epsilon: 10
+#      reg: 0.00001
+#      scaling: log
+#      epochs: 20
+    external.EASER:
+      meta:
+        verbose: True
+        save_recs: True
+        hyper_max_evals: 1
+        hyper_opt_alg: tpe
+      l2_norm: 1320
+    external.RP3beta: #from TOIS
+      meta:
+        hyper_max_evals: 1
+        hyper_opt_alg: tpe
+        verbose: True
+        save_recs: True
+      neighborhood: 546
+      alpha: 1.0807
+      beta: 0.7029
+      normalize_similarity: True