Set arguments in order to reproduce 'master' performances with Neural setting

main.py

@@ -15,9 +15,21 @@ def main(args):
     print('Running generalized funnelling...')
 
     data = MultilingualDataset.load(args.dataset)
-    # data.set_view(languages=['it', 'da'])
+    data.set_view(languages=['it', 'da'])
     data.show_dimensions()
     lX, ly = data.training()
+
+    # Testing zero shot experiments
+    # zero_shot_setting = True
+    # if zero_shot_setting:
+    #     # _lX = {}
+    #     _ly = {}
+    #     train_langs = ['it']
+    #     for train_lang in train_langs:
+    #         # _lX[train_lang] = lX[train_lang]
+    #         _ly[train_lang] = ly[train_lang]
+    #     ly = _ly
+
     lXte, lyte = data.test()
 
     # Init multilingualIndex - mandatory when deploying Neural View Generators...
@@ -33,7 +45,7 @@ def main(args):
         embedder_list.append(posteriorEmbedder)
 
     if args.muse_embedder:
-        museEmbedder = MuseGen(muse_dir=args.muse_dir, n_jobs=args.n_jobs)
+        museEmbedder = MuseGen(muse_dir=args.muse_dir, n_jobs=args.n_jobs, zero_shot=True)
         embedder_list.append(museEmbedder)
 
     if args.wce_embedder:
@@ -99,7 +111,7 @@ def main(args):
                             microf1=microf1,
                             macrok=macrok,
                             microk=microk,
-                            notes='')
+                            notes=f'Train langs: {sorted(lX.keys())}')
     print('Averages: MF1, mF1, MK, mK', np.round(np.mean(np.array(metrics), axis=0), 3))
 
     overall_time = round(time.time() - time_init, 3)
@@ -112,8 +124,8 @@ if __name__ == '__main__':
     parser.add_argument('dataset', help='Path to the dataset')
 
     parser.add_argument('-o', '--output', dest='csv_dir', metavar='',
-                        help='Result file (default ../csv_logs/gfun/gfun_results.csv)', type=str,
-                        default='../csv_logs/gfun/gfun_results.csv')
+                        help='Result file (default csv_logs/gfun/gfun_results.csv)', type=str,
+                        default='csv_logs/gfun/gfun_results.csv')
 
     parser.add_argument('-x', '--post_embedder', dest='post_embedder', action='store_true',
                         help='deploy posterior probabilities embedder to compute document embeddings',

run.sh

@@ -1,6 +1,8 @@
 #!/usr/bin/env bash
 
-python main.py /home/moreo/CLESA/rcv2/rcv1-2_doclist_trByLang1000_teByLang1000_processed_run0.pickle -g --gpus 0
+echo Running Zero-shot experiments [output at csv_logs/gfun/zero_shot_gfun.csv]
+
+python main.py /home/moreo/CLESA/rcv2/rcv1-2_doclist_trByLang1000_teByLang1000_processed_run0.pickle -m -o csv_logs/gfun/zero_shot_gfun.csv --gpus 0
 
 #for i in {0..10..1}
 #do

src/util/common.py

@@ -378,7 +378,7 @@ def get_method_name(args):
     for i, conf in enumerate(_id_conf):
         if conf:
             _id += _id_name[i]
-    _id = _id if not args.gru_wce else _id + '_wce'
+    _id = _id if not args.rnn_wce else _id + '_wce'
     _dataset_path = args.dataset.split('/')[-1].split('_')
     dataset_id = _dataset_path[0] + _dataset_path[-1]
     return _id, dataset_id

src/view_generators.py

@@ -99,7 +99,7 @@ class MuseGen(ViewGen):
     View Generator (m): generates document representation via MUSE embeddings (Fasttext multilingual word
     embeddings). Document embeddings are obtained via weighted sum of document's constituent embeddings.
     """
-    def __init__(self, muse_dir='../embeddings', n_jobs=-1):
+    def __init__(self, muse_dir='../embeddings', zero_shot=False, n_jobs=-1):
         """
         Init the MuseGen.
         :param muse_dir: string, path to folder containing muse embeddings
@@ -111,6 +111,7 @@ class MuseGen(ViewGen):
         self.langs = None
         self.lMuse = None
         self.vectorizer = TfidfVectorizerMultilingual(sublinear_tf=True, use_idf=True)
+        self.zero_shot = zero_shot
 
     def fit(self, lX, ly):
         """
@@ -135,16 +136,34 @@ class MuseGen(ViewGen):
         :param lX: dict {lang: indexed documents}
         :return: document projection to the common latent space.
         """
-        lX = self.vectorizer.transform(lX)
+        # Testing zero-shot experiments
+        if self.zero_shot:
+            lX = {l: self.vectorizer.vectorizer[l].transform(lX[l]) for l in self.langs if lX[l] is not None}
+        else:
+            lX = self.vectorizer.transform(lX)
         XdotMUSE = Parallel(n_jobs=self.n_jobs)(
-            delayed(XdotM)(lX[lang], self.lMuse[lang], sif=True) for lang in self.langs)
-        lZ = {lang: XdotMUSE[i] for i, lang in enumerate(self.langs)}
+            delayed(XdotM)(lX[lang], self.lMuse[lang], sif=True) for lang in sorted(lX.keys()))
+        lZ = {lang: XdotMUSE[i] for i, lang in enumerate(sorted(lX.keys()))}
         lZ = _normalize(lZ, l2=True)
         return lZ
 
     def fit_transform(self, lX, ly):
+        print('## NB: Calling fit_transform!')
+        if self.zero_shot:
+            return self.fit(lX, ly).transform(self.zero_shot_experiments(lX))
         return self.fit(lX, ly).transform(lX)
 
+    def zero_shot_experiments(self, lX, train_langs: list = ['it']):
+        print(f'# Zero-shot setting! Training langs will be set to: {sorted(train_langs)}')
+        _lX = {}
+        for lang in self.langs:
+            if lang in train_langs:
+                _lX[lang] = lX[lang]
+            else:
+                _lX[lang] = None
+        lX = _lX
+        return lX
+
 
 class WordClassGen(ViewGen):
     """