passing NeuralLDA.train_model(top_words=10) to AVITM_model.train_model(top_words=10)

octis/models/CTM.py

@@ -134,7 +134,7 @@ def train_model(self, dataset, hyperparameters=None, top_words=10):
                 topic_prior_variance=self.hyperparameters["prior_variance"],
                 top_words=top_words)
 
-            self.model.fit(x_train, x_valid, verbose=False)
+            self.model.fit(x_train, x_valid, verbose=self.hyperparameters["verbose"], save_dir=self.hyperparameters["save_dir"])
             result = self.inference(x_test)
             return result
 
@@ -161,7 +161,7 @@ def train_model(self, dataset, hyperparameters=None, top_words=10):
              topic_prior_variance=self.hyperparameters["prior_variance"],
              top_words=top_words)
 
-        self.model.fit(x_train, None, verbose=False)
+        self.model.fit(x_train, None, verbose=self.hyperparameters["verbose"], save_dir=self.hyperparameters["save_dir"])
         result = self.model.get_info()
         return result
 

octis/models/contextualized_topic_models/models/ctm.py

@@ -273,7 +273,7 @@ def fit(self, train_dataset, validation_dataset=None,
 
         train_loader = DataLoader(
             self.train_data, batch_size=self.batch_size, shuffle=True,
-            num_workers=self.num_data_loader_workers)
+            num_workers=self.num_data_loader_workers, drop_last=True)
 
         # init training variables
         train_loss = 0
@@ -301,7 +301,7 @@ def fit(self, train_dataset, validation_dataset=None,
             if self.validation_data is not None:
                 validation_loader = DataLoader(
                     self.validation_data, batch_size=self.batch_size,
-                    shuffle=True, num_workers=self.num_data_loader_workers)
+                    shuffle=True, num_workers=self.num_data_loader_workers, drop_last=True)
                 # train epoch
                 s = datetime.datetime.now()
                 val_samples_processed, val_loss = self._validation(

octis/models/pytorchavitm/AVITM.py

@@ -98,14 +98,14 @@ def train_model(self, dataset, hyperparameters=None, top_words=10):
             solver=self.hyperparameters['solver'], num_epochs=self.hyperparameters['num_epochs'],
             reduce_on_plateau=self.hyperparameters['reduce_on_plateau'], num_samples=self.hyperparameters[
                 'num_samples'], topic_prior_mean=self.hyperparameters["prior_mean"],
-            topic_prior_variance=self.hyperparameters["prior_variance"]
+            topic_prior_variance=self.hyperparameters["prior_variance"], verbose=self.hyperparameters["verbose"], top_words=top_words,
         )
 
         if self.use_partitions:
-            self.model.fit(x_train, x_valid)
+            self.model.fit(x_train, x_valid, save_dir=self.hyperparameters["save_dir"])
             result = self.inference(x_test)
         else:
-            self.model.fit(x_train, None)
+            self.model.fit(x_train, None, save_dir=self.hyperparameters["save_dir"])
             result = self.model.get_info()
         return result
 

octis/models/pytorchavitm/avitm/avitm_model.py

@@ -19,7 +19,7 @@ class AVITM_model(object):
     def __init__(self, input_size, num_topics=10, model_type='prodLDA', hidden_sizes=(100, 100),
                  activation='softplus', dropout=0.2, learn_priors=True, batch_size=64, lr=2e-3, momentum=0.99,
                  solver='adam', num_epochs=100, reduce_on_plateau=False, topic_prior_mean=0.0,
-                 topic_prior_variance=None, num_samples=10, num_data_loader_workers=0, verbose=False):
+                 topic_prior_variance=None, num_samples=10, num_data_loader_workers=0, verbose=False, top_words=10):
         """
         Initialize AVITM model.
 
@@ -68,6 +68,7 @@ def __init__(self, input_size, num_topics=10, model_type='prodLDA', hidden_sizes
         # assert isinstance(topic_prior_variance, float), \
         #    "topic prior_variance must be type float"
 
+        self.top_words = top_words
         self.input_size = input_size
         self.num_topics = num_topics
         self.verbose = verbose
@@ -240,7 +241,7 @@ def fit(self, train_dataset, validation_dataset, save_dir=None):
         self.validation_data = validation_dataset
         train_loader = DataLoader(
             self.train_data, batch_size=self.batch_size, shuffle=True,
-            num_workers=self.num_data_loader_workers)
+            num_workers=self.num_data_loader_workers, drop_last=True)
 
         # init training variables
         train_loss = 0
@@ -267,7 +268,7 @@ def fit(self, train_dataset, validation_dataset, save_dir=None):
             if self.validation_data is not None:
                 validation_loader = DataLoader(
                     self.validation_data, batch_size=self.batch_size, shuffle=True,
-                    num_workers=self.num_data_loader_workers)
+                    num_workers=self.num_data_loader_workers, drop_last=True)
                 # train epoch
                 s = datetime.datetime.now()
                 val_samples_processed, val_loss = self._validation(validation_loader)
@@ -347,7 +348,7 @@ def get_topics(self, k=10):
 
     def get_info(self):
         info = {}
-        topic_word = self.get_topics()
+        topic_word = self.get_topics(k=self.top_words) # or self.input_size
         topic_word_dist = self.get_topic_word_mat()
         # topic_document_dist = self.get_topic_document_mat()
         info['topics'] = topic_word