Final refinements

README.md

-# NLP Machine Learning Project (Machine Translator EN <-> DE)
+# NLP Machine Learning Project | Machine Translator EN -> DE
 
 ## BLEU scores
 
@@ -20,3 +20,6 @@ Train a machine translator based on a recurrent neural network.
 3. Run `conda env create -f env.yaml`
 4. Run `conda activate nlp-machine-learning-project`
 5. Set the created environment as the python interpreter within your IDE
+6. Run the python files within the models directory
+
+link to the repo: https://git.cs.uni-paderborn.de/marvnsch/nlp-machine-translation-project
\ No newline at end of file

models/RNN_no_attention_bidirectional/RNN_no_attention_bidirectional.py

@@ -103,8 +103,7 @@ class Seq2Seq(nn.Module):
 # setup environment
 source_data_path = str(project_root / "data/tokenizer-data/news-commentary-v11.en")
 target_data_path = str(project_root / "data/tokenizer-data/news-commentary-v11.de")
-#device = utils.pytorch.get_available_device()
-device = torch.device('cpu')
+device = utils.pytorch.get_available_device()
 date_time_now = datetime.now().strftime("%m%d%Y_%H%M")
 model_output_path = str(work_dir / f"./checkpoints/{date_time_now}_RNN_no_attention_bidirectional.pt")
 
@@ -164,8 +163,7 @@ data_loader = data.preprocessing.data_loader(source=source_data,
                                              target_tokenizer=target_tokenizer,
                                              dataset_size=dataset_size,
                                              torch_device=device,
-                                             data_split=train_dev_val_split,
-                                             return_data=True)
+                                             data_split=train_dev_val_split)
 
 source_data = None
 target_data = None
@@ -178,21 +176,20 @@ except FileExistsError:
 
 
 # train the model
-#utils.operation.train(model=model,
-#                      data_loader=data_loader,
-#                      criterion=criterion,
-#                      optimizer=optimizer,
-#                      num_epochs=num_epochs,
-#                      num_of_batches_per_epoch=train_batches_count,
-#                      saving_interval=500,
-#                      model_output_path=model_output_path)
+utils.operation.train(model=model,
+                      data_loader=data_loader,
+                      criterion=criterion,
+                      optimizer=optimizer,
+                      num_epochs=num_epochs,
+                      num_of_batches_per_epoch=train_batches_count,
+                      saving_interval=500,
+                      model_output_path=model_output_path)
 
 _, _, val_data_loader = data_loader()
 
 # evaluate the model
-utils.operation.evaluateModelFromFile(model=model,
-                                      data_loader=data_loader,
-                                      model_path=str(work_dir / "./checkpoints/01242024_1659_RNN_no_attention_bidirectional.pth"),
+utils.operation.evaluate(model=model,
+                         val_data_loader=val_data_loader,
                          source_tokenizer=source_tokenizer,
                          target_tokenizer=target_tokenizer)
 

models/RNN_no_attention_unidirectional/RNN_no_attention_unidirectional.py

@@ -157,8 +157,7 @@ data_loader = data.preprocessing.data_loader(source=source_data,
                                              target_tokenizer=target_tokenizer,
                                              dataset_size=dataset_size,
                                              torch_device=device,
-                                             data_split=train_dev_val_split,
-                                             return_data=True)
+                                             data_split=train_dev_val_split)
 
 source_data = None
 target_data = None
@@ -171,20 +170,19 @@ except FileExistsError:
 
 
 # train the model
-#utils.operation.train(model=model,
-#                      data_loader=data_loader,
-#                      criterion=criterion,
-#                      optimizer=optimizer,
-#                      num_epochs=num_epochs,
-#                      num_of_batches_per_epoch=train_batches_count,
-#                      saving_interval=500,
-#                      model_output_path=model_output_path)
+utils.operation.train(model=model,
+                      data_loader=data_loader,
+                      criterion=criterion,
+                      optimizer=optimizer,
+                      num_epochs=num_epochs,
+                      num_of_batches_per_epoch=train_batches_count,
+                      saving_interval=500,
+                      model_output_path=model_output_path)
 
 _, _, val_data_loader = data_loader()
 
 # evaluate the model
-utils.operation.evaluateModelFromFile(model=model,
-                                      data_loader=data_loader,
-                                      model_path=str(work_dir / "./checkpoints/01242024_1324_RNN_no_attention_unidirectional.pth"),
+utils.operation.evaluate(model=model,
+                         val_data_loader=val_data_loader,
                          source_tokenizer=source_tokenizer,
                          target_tokenizer=target_tokenizer)
\ No newline at end of file

models/RNN_with_attention_bidirectional/RNN_with_attention_bidirectional.py

@@ -196,8 +196,7 @@ data_loader = data.preprocessing.data_loader(source=source_data,
                                              target_tokenizer=target_tokenizer,
                                              dataset_size=dataset_size,
                                              torch_device=device,
-                                             data_split=train_dev_val_split,
-                                             return_data=True)
+                                             data_split=train_dev_val_split)
 
 source_data = None
 target_data = None
@@ -209,22 +208,21 @@ except FileExistsError:
     pass
 
 # train the model
-#utils.operation.train(model=model,
-#                      data_loader=data_loader,
-#                      criterion=criterion,
-#                      optimizer=optimizer,
-#                      num_epochs=num_epochs,
-#                      num_of_batches_per_epoch=train_batches_count,
-#                      saving_interval=500,
-#                      model_output_path=model_output_path)
+utils.operation.train(model=model,
+                      data_loader=data_loader,
+                      criterion=criterion,
+                      optimizer=optimizer,
+                      num_epochs=num_epochs,
+                      num_of_batches_per_epoch=train_batches_count,
+                      saving_interval=500,
+                      model_output_path=model_output_path)
 
 
 _, _, val_data_loader = data_loader()
 
 # evaluate the model
-utils.operation.evaluateModelFromFile(model=model,
-                                      data_loader=data_loader,
-                                      model_path=str(work_dir / "./checkpoints/01242024_1748_RNN_attention_bidirectional.pth"),
+utils.operation.evaluate(model=model,
+                         val_data_loader=val_data_loader,
                          source_tokenizer=source_tokenizer,
                          target_tokenizer=target_tokenizer)
 

utils/operation.py

@@ -101,7 +101,7 @@ def train(model, data_loader, num_of_batches_per_epoch: int,
 def evaluate(model, val_data_loader,
              source_tokenizer: Tokenizer,
              target_tokenizer: Tokenizer,
-             print_num_examples: int = 0):
+             print_num_examples: int = 10):
     model.eval()
 
     source_corpus = []
@@ -127,8 +127,6 @@ def evaluate(model, val_data_loader,
         print(f"< {candidate_corpus[idx]}")
         print()
 
-    print(f"BLEU score result: {bleu_score(candidate_corpus, reference_corpus)}")
-
 
 def evaluateModelFromFile(model: torch.nn.Module, data_loader, model_path:str,
                           source_tokenizer: Tokenizer,