Merge branch 'master' into 'kreuzer-master-patch-32167'

# Conflicts:
#   scripts/ActorOnly+CEL/main_ActorOnly.py
#   scripts/ActorOnly+CEL/main_CrossEntropy.py

main_preprocess.py

+import torch
+import stanza
+from gensim.models import KeyedVectors
+from datasets import load_dataset
+from structures import *
+
+
+# loads skipgram gensim
+file_name = "data/1-billion-word-language-modeling-benchmark-r13output.word2vec.vec"
+model_gensim = KeyedVectors.load_word2vec_format(file_name)
+
+# initialize tokenizer, => sentences splitting and tokenizing, !pip install stanza 
+nlp = stanza.Pipeline(lang='en', processors='tokenize')
+
+# loads dataset cnn_dailymail, !pip install datasets
+dataset = load_dataset('ccdv/cnn_dailymail', '3.0.0', split='test[:5]') # extract subset for testing
+# update path 
+x = PreprocessedDataSet('test5', dataset, model_gensim, nlp)
+

models.py

@@ -3,6 +3,8 @@ import torch
 from torch import nn
 import numpy as np
 
+import utils #
+
 
 class SummarisationModel(nn.Module):
 

plot_training_curves.py

-
 import torch
 import numpy as np
 import matplotlib.pyplot as plt
+import pathlib
+
+hist_path = pathlib.Path('hist')
+actor_only_hist = torch.load(hist_path/'model_actor_only_val_rouge_hist.pt')
+CE_hist = torch.load(hist_path/'model_CE_val_rouge_hist.pt')
 
-hist = torch.rand(10)
-scratch_hist = torch.rand(10)
+ahist = torch.Tensor(actor_only_hist)
+chist = torch.Tensor(CE_hist)
 
-num_epochs = 10
+num_epochs = 2
 
 ohist = []
 shist = []
 
-ohist = [h.cpu().numpy() for h in hist]
-shist = [h.cpu().numpy() for h in scratch_hist]
+ohist = [h.cpu().numpy() for h in ahist]
+shist = [h.cpu().numpy() for h in chist]
 
-plt.title("Validation Accuracy vs. Number of Training Epochs")
+plt.title("Validation Rouge vs. Number of Training Epochs")
 plt.xlabel("Training Epochs")
-plt.ylabel("Validation Accuracy")
-plt.plot(range(1,num_epochs+1),ohist,label="Pretrained")
-plt.plot(range(1,num_epochs+1),shist,label="Scratch")
-plt.ylim((0,1.))
+plt.ylabel("Validation Rouge")
+plt.plot(range(1,num_epochs+1),ohist,label="Actor Only")
+plt.plot(range(1,num_epochs+1),shist,label="Cross Entropy")
+plt.ylim((min(actor_only_hist + CE_hist), max(actor_only_hist + CE_hist)))
 plt.xticks(np.arange(1, num_epochs+1, 1.0))
 plt.legend()
 plt.show()
+

scripts/ActorOnly+CEL/main_ActorOnly.py

@@ -3,27 +3,28 @@ from torch import nn
 import numpy as np
 from structures import *
 from models import *
-import utils
+import time
 
-# loads dataset cnn_dailymail
-train_dataset = PreprocessedDataSet('/workspace/students/kreuzer/train')
-validation_dataset = PreprocessedDataSet('/workspace/students/kreuzer/val')
-test_dataset = PreprocessedDataSet('/workspace/students/kreuzer/test')
+# loads dataset after preprocessing
+train_data = PreprocessedDataSet('/workspace/students/kreuzer/train')
+val_data = PreprocessedDataSet('/workspace/students/kreuzer/val')
+test_data = PreprocessedDataSet('/workspace/students/kreuzer/test')
 
 # hyperparameters
-epochs=20
-batch_size=20
+epochs=20 # 20
+batch_size=20 # 20
 learning_rate=0.001
 
-training_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, collate_fn=lambda x:x)
+train_dataloader = torch.utils.data.DataLoader(train_data, batch_size=batch_size, shuffle=True, collate_fn = lambda x:x) # collate_fn 
+#test_dataloader = torch.utils.data.DataLoader(test_data, batch_size=batch_size, shuffle=True, collate_fn = lambda x: x)
 
-m = ActorOnlySummarisation()
+m = ActorOnlySummarisationModel() # tip error fix!
 
 since = time.time()
 val_rouge_history = []
 
 best_rouge = 0.0
-best_model_wts = copy.deepcopy(m.state_dict())
+best_model_wts = copy.deepcopy(m.state_dict()) 
 
 for epoch in range(epochs):
 
@@ -32,12 +33,13 @@ for epoch in range(epochs):
     print('-' * 10)
 
     # train phase
-    epoch_loss, epoch_rouge = m.training_epoch(training_dataloader)
+    epoch_loss, epoch_rouge = m.training_epoch(train_dataloader) # train_dataloader
     print('Train Loss: {:.4f} Rouge Score: {:.4f}'.format(epoch_loss, epoch_rouge))
     
     # validation phase 
-    epoch_rouge = m.validation(validation_dataset)
-    val_rouge_history.append(epoch_rouge)
+    epoch_rouge = m.validation(val_data) #
+    val_rouge_history.append(epoch_rouge)    
+
     print('Validation Rouge Score: {:.4f}'.format(epoch_rouge))
 
     # epoch completed, deep copy the best model sofar
@@ -51,11 +53,9 @@ print('Training + validation complete in {:.0f}m {:.0f}s'.format(time_elapsed //
 print('Best val rouge: {:4f}'.format(best_rouge))
 
 # write val_rouge_history in file
-with open("ActorOnly_val_history.txt", "a") as f:
-    f.write(">>>Start\n")
-    for event in val_rouge_history:
-        f.write(str(event)+"\n")
-    f.write("\n")
+pathlib.Path('hist').mkdir(parents=True, exist_ok=True)
+hist_path = pathlib.Path('hist')
+torch.save(val_rouge_history, hist_path/'model_actor_only_val_rouge_hist.pt') # all hist in one folder hist?
 
 # load best model weights
 m.load_state_dict(best_model_wts)
@@ -63,15 +63,14 @@ m.load_state_dict(best_model_wts)
 # for Critic: save model_actor_only   
 torch.save(m.state_dict(), 'model_actor_only_wts.pth')
 # set sent_vecs 
-train_dataset.compute_sent_vecs(m)
-validation_dataset.compute_sent_vecs(m)
-test_dataset.compute_sent_vecs(m)
+train_data.compute_sent_vecs(m)  #
+val_data.compute_sent_vecs(m)
+test_data.compute_sent_vecs(m)
 
 # testing
-
 since = time.time()
 
-epoch_rouge_1, epoch_rouge_2, epoch_rouge_l = m.test(test_dataset)
+epoch_rouge_1, epoch_rouge_2, epoch_rouge_l = m.test(test_data)
 print('Test rouge_1: {:.4f} rouge_2: {:.4f} rouge_l: {:.4f} mean: {:.4f}'.format(epoch_rouge_1, epoch_rouge_2, epoch_rouge_l, (epoch_rouge_1+epoch_rouge_2+epoch_rouge_l)/3.0))
 
 # after testing completed

scripts/ActorOnly+CEL/main_CrossEntropy.py

@@ -3,22 +3,23 @@ from torch import nn
 import numpy as np
 from structures import *
 from models import *
-import utils
+import time
 
 # loads dataset cnn_dailymail
-train_dataset = PreprocessedDataSet('/workspace/students/kreuzer/train')
-validation_dataset = PreprocessedDataSet('/workspace/students/kreuzer/val')
-test_dataset = PreprocessedDataSet('/workspace/students/kreuzer/test')
+train_data = PreprocessedDataSet('/workspace/students/kreuzer/train')
+validation_data = PreprocessedDataSet('/workspace/students/kreuzer/val')
+test_data = PreprocessedDataSet('/workspace/students/kreuzer/test')
 
 # hyperparameters
 epochs=20
-batch_size=20
+batch_size=50 # 5
 learning_rate=0.001
 
-training_dataloader = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, collate_fn=lambda x:x)
+train_dataloader = torch.utils.data.DataLoader(train_data, batch_size=batch_size, shuffle=True, collate_fn=lambda x:x)
 
 m = SummarisationModelWithCrossEntropyLoss()
 
+# training
 since = time.time()
 val_rouge_history = []
 
@@ -27,7 +28,6 @@ best_model_wts = copy.deepcopy(m.state_dict())
 
 for epoch in range(epochs):
     
-    print()
     print('Epoch {}/{}'.format(epoch+1, epochs))
     print('-' * 10)
 
@@ -36,7 +36,8 @@ for epoch in range(epochs):
     print('Train Loss: {:.4f}'.format(epoch_loss))
     
     # validation phase 
-    epoch_rouge = m.validation(validation_dataset)
+    epoch_rouge = m.validation(val_data)
+
     val_rouge_history.append(epoch_rouge)    
     print('Validation Rouge Score: {:.4f}'.format(epoch_rouge))
 
@@ -51,11 +52,9 @@ print('Training + validation complete in {:.0f}m {:.0f}s'.format(time_elapsed //
 print('Best val rouge: {:4f}'.format(best_rouge))
 
 # write val_rouge_history in file
-with open("CEL_val_history.txt", "a") as f:
-    f.write(">>>Start\n")
-    for event in val_rouge_history:
-        f.write(str(event)+"\n")
-    f.write("\n")
+pathlib.Path('hist').mkdir(parents=True, exist_ok=True)
+hist_path = pathlib.Path('hist')
+torch.save(val_rouge_history, hist_path/'model_CE_val_rouge_hist.pt')
 
 # load best model weights
 m.load_state_dict(best_model_wts)
@@ -63,7 +62,8 @@ m.load_state_dict(best_model_wts)
 # testing
 since = time.time()
 
-epoch_rouge_1, epoch_rouge_2, epoch_rouge_l = m.test(test_dataset)
+epoch_rouge_1, epoch_rouge_2, epoch_rouge_l = m.test(test_data)
+
 print('Test rouge_1: {:.4f} rouge_2: {:.4f} rouge_l: {:.4f} mean: {:.4f}'.format(epoch_rouge_1, epoch_rouge_2, epoch_rouge_l, (epoch_rouge_1+epoch_rouge_2+epoch_rouge_l)/3.0))
 
 # after testing completed