similarity saving added

data/twitter/config.py

@@ -38,13 +38,13 @@ class TwitterConfig(Config):
     attention_weight_decay = 0.05
     classification_weight_decay = 0.05
 
-    device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
+    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-    image_model_name = '../../../../../media/external_10TB/10TB/ghorbanpoor/vit-base-patch16-224'
+    image_model_name = '../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/vit-base-patch16-224'
     image_embedding = 768
-    text_encoder_model = "../../../../../media/external_10TB/10TB/ghorbanpoor/bert-base-uncased"
+    text_encoder_model = "../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/bert-base-uncased"
     # text_encoder_model = "/home/faeze/PycharmProjects/new_fake_news_detectioin/bert/bert-base-uncased"
-    text_tokenizer = "../../../../../media/external_10TB/10TB/ghorbanpoor/bert-base-uncased"
+    text_tokenizer = "../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/bert-base-uncased"
     # text_tokenizer = "/home/faeze/PycharmProjects/new_fake_news_detectioin/bert/bert-base-uncased"
     text_embedding = 768
     max_length = 32
@@ -65,9 +65,9 @@ class TwitterConfig(Config):
         self.classification_lr = trial.suggest_loguniform('classification_lr', 1e-5, 1e-1)
 
         self.head_weight_decay = trial.suggest_loguniform('head_weight_decay', 1e-5, 1e-1)
-        # self.attention_weight_decay = trial.suggest_loguniform('attention_weight_decay', 1e-5, 1e-1)
-        self.classification_weight_decay = trial.suggest_loguniform('classification_weight_decay', 1e-5, 1e-1)
+        self.attention_weight_decay = trial.suggest_loguniform('attention_weight_decay', 1e-5, 1e-1)
+        # self.classification_weight_decay = trial.suggest_loguniform('classification_weight_decay', 1e-5, 1e-1)
 
-        # self.projection_size = trial.suggest_categorical('projection_size', [256, 128, 64])
+        self.projection_size = trial.suggest_categorical('projection_size', [256, 128, 64])
         # self.hidden_size = trial.suggest_categorical('hidden_size', [256, 128, 64, ])
-        # self.dropout = trial.suggest_categorical('drop_out', [0.1, 0.3, 0.5, ])
+        self.dropout = trial.suggest_categorical('drop_out', [0.1, 0.3, 0.5, ])

data/weibo/config.py

@@ -33,13 +33,13 @@ class WeiboConfig(Config):
     projection_size = 64
     dropout = 0.5
 
-    device = torch.device("cuda:1" if torch.cuda.is_available() else "cpu")
+    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
-    image_model_name = '../../../../../media/external_10TB/10TB/ghorbanpoor/vit-base-patch16-224'
+    image_model_name = '../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/vit-base-patch16-224'
     image_embedding = 768
-    text_encoder_model = "../../../../../media/external_10TB/10TB/ghorbanpoor/bert-base-chinese"
+    text_encoder_model = "../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/bert-base-chinese"
     # text_encoder_model = "/home/faeze/PycharmProjects/new_fake_news_detectioin/bert/bert-base-uncased"
-    text_tokenizer = "../../../../../media/external_10TB/10TB/ghorbanpoor/bert-base-chinese"
+    text_tokenizer = "../../../../../media/external_10TB/10TB/ghorbanpoor/huggingface/bert-base-chinese"
     # text_tokenizer = "/home/faeze/PycharmProjects/new_fake_news_detectioin/bert/bert-base-uncased"
     text_embedding = 768
     max_length = 200
@@ -59,9 +59,9 @@ class WeiboConfig(Config):
         self.classification_lr = trial.suggest_loguniform('classification_lr', 1e-5, 1e-1)
 
         self.head_weight_decay = trial.suggest_loguniform('head_weight_decay', 1e-5, 1e-1)
-        self.classification_weight_decay = trial.suggest_loguniform('classification_weight_decay', 1e-5, 1e-1)
+        # self.classification_weight_decay = trial.suggest_loguniform('classification_weight_decay', 1e-5, 1e-1)
 
         self.projection_size = trial.suggest_categorical('projection_size', [256, 128, 64])
-        self.hidden_size = trial.suggest_categorical('hidden_size', [256, 128, 64, ])
+        # self.hidden_size = trial.suggest_categorical('hidden_size', [256, 128, 64, ])
         self.dropout = trial.suggest_categorical('drop_out', [0.1, 0.3, 0.5, ])
 

model.py

@@ -4,6 +4,7 @@ from torch import nn
 
 from image import ImageTransformerEncoder, ImageResnetEncoder
 from text import TextEncoder
+import numpy as np
 
 
 class ProjectionHead(nn.Module):
@@ -18,7 +19,7 @@ class ProjectionHead(nn.Module):
         self.gelu = nn.GELU()
         self.fc = nn.Linear(config.projection_size, config.projection_size)
         self.dropout = nn.Dropout(config.dropout)
-        self.layer_norm = nn.LayerNorm(config.projection_size)
+        self.layer_norm = nn.BatchNorm1d(config.projection_size)
 
     def forward(self, x):
         projected = self.projection(x)
@@ -81,39 +82,36 @@ class FakeNewsModel(nn.Module):
         self.text_embeddings = self.text_projection(text_features)
 
         self.multimodal_embeddings = torch.cat((self.image_embeddings, self.text_embeddings), dim=1)
-        self.logits = (self.text_embeddings @ self.image_embeddings.T)
 
         score = self.classifier(self.multimodal_embeddings)
         probs, output = torch.max(score.data, dim=1)
 
-        return output, score
+        self.similarity = self.text_embeddings @ self.image_embeddings.T
 
+        return output, score
 
 
 def calculate_loss(model, score, label):
-    similarity = calculate_similarity_loss(model.config, model.image_embeddings, model.text_embeddings)
-    # fake = (label == 1).nonzero()
-    # real = (label == 0).nonzero()
-    # s_loss = 0 * similarity[fake].mean() + similarity[real].mean()
-    s_loss = similarity.mean()
+    s_loss = calculate_similarity_loss(model.image_embeddings, model.text_embeddings)
     c_loss = model.classifier_loss_function(score, label)
     loss = model.config.loss_weight * c_loss + s_loss
     return loss, c_loss, s_loss
 
 
-def calculate_similarity_loss(config, image_embeddings, text_embeddings):
-    # Calculating the Loss
+def calculate_similarity_loss(image_embeddings, text_embeddings):
     logits = (text_embeddings @ image_embeddings.T)
-    images_similarity = image_embeddings @ image_embeddings.T
-    texts_similarity = text_embeddings @ text_embeddings.T
+    images_similarity = (image_embeddings @ image_embeddings.T)
+    texts_similarity = (text_embeddings @ text_embeddings.T)
     targets = F.softmax((images_similarity + texts_similarity) / 2, dim=-1)
-    texts_loss = cross_entropy(logits, targets, reduction='none')
-    images_loss = cross_entropy(logits.T, targets.T, reduction='none')
+    texts_loss = cross_entropy(logits, targets, reduction='mean')
+    images_loss = cross_entropy(logits.T, targets.T, reduction='mean')
     loss = (images_loss + texts_loss) / 2.0
     return loss
 
 
 def cross_entropy(preds, targets, reduction='none'):
+    # entropy = torch.nn.CrossEntropyLoss(reduction=reduction)
+    # return entropy(preds, targets)
     log_softmax = nn.LogSoftmax(dim=-1)
     loss = (-targets * log_softmax(preds)).sum(1)
     if reduction == "none":

test_main.py

@@ -13,13 +13,11 @@ from model import FakeNewsModel, calculate_loss
 
 
 def test(config, test_loader, trial_number=None):
-    if trial_number:
-        try:
-            checkpoint = torch.load(str(config.output_path) + '/checkpoint_' + str(trial_number) + '.pt')
-        except:
-            checkpoint = torch.load(str(config.output_path) + '/checkpoint.pt')
-    else:
-        checkpoint = torch.load(str(config.output_path) + '/checkpoint.pt', map_location=torch.device(config.device))
+
+    try:
+        checkpoint = torch.load(str(config.output_path) + '/checkpoint_' + str(trial_number) + '.pt')
+    except:
+        checkpoint = torch.load(str(config.output_path) + '/checkpoint.pt')
 
     try:
         parameters = checkpoint['parameters']
@@ -49,6 +47,7 @@ def test(config, test_loader, trial_number=None):
     scores = []
     ids = []
     losses = []
+    similarities = []
     tqdm_object = tqdm(test_loader, total=len(test_loader))
     for i, batch in enumerate(tqdm_object):
         batch = batch_constructor(config, batch)
@@ -64,6 +63,7 @@ def test(config, test_loader, trial_number=None):
             text_features.append(model.text_embeddings.detach())
             multimodal_features.append(model.multimodal_embeddings.detach())
             concat_features.append(model.classifier.embeddings.detach())
+            similarities.append(model.similarity.detach())
             losses.append((loss.detach(), c_loss.detach(), s_loss.detach()))
 
     s = ''
@@ -79,6 +79,7 @@ def test(config, test_loader, trial_number=None):
     save_embedding(text_features, fname=str(config.output_path) + '/new_text_features.tsv')
     save_embedding(multimodal_features, fname=str(config.output_path) + '/new_multimodal_features_.tsv')
     save_embedding(concat_features, fname=str(config.output_path) + '/new_concat_features.tsv')
+    save_embedding(similarities, fname=str(config.output_path) + '/similarities.tsv')
 
     config_parameters = str(config)
     with open(config.output_path + '/new_parameters.txt', 'w') as f: