Update recommendation.py

recommendation.py

@@ -3,6 +3,7 @@ from sentence_transformers import SentenceTransformer
 from torchvision import transforms
 from models.encoder import Encoder
 from indexer import Indexer
+import numpy as np
 import torch
 import os
 
@@ -18,22 +19,33 @@ dataset = concatenate_datasets([dataset['train'], dataset['test']])
 latent_data = torch.load("data/latent_data_final.bin")
 embeddings = torch.load("data/embeddings.bin")
 
-text_embedding_index = Indexer(embeddings)
-image_embedding_index = Indexer(latent_data)
+
+def row_wise_normalize_and_concatenate(array1, array2):
+    normalized_array1 = array1 / np.linalg.norm(array1, axis=1, keepdims=True)
+    normalized_array2 = array2 / np.linalg.norm(array2, axis=1, keepdims=True)
+
+    concatenated_array = np.concatenate((normalized_array1, normalized_array2), axis=0)
+
+    return concatenated_array
+
+result_array = row_wise_normalize_and_concatenate(latent_data, embeddings)
+
+index = Indexer(result_array)
 
 def get_recommendations(image, title, k):
-#   title = [dataset[product_id]['title']]
+  title = [dataset[product_id]['title']]
   title_embeds = model.encode([title], normalize_embeddings=True)
   image = transforms.ToTensor()(image.convert("L"))
   image_embeds =  encoder(image).detach().numpy()
+  final_embeds = np.concatenate((image_embeds,title_embeds), axis=0)
 
-  image_candidates = image_embedding_index.topk(image_embeds,k=k)
-  title_candidates = text_embedding_index.topk(title_embeds, k=k)
+  candidates = image_embedding_index.topk(final_embeds,k=k)
+  # title_candidates = text_embedding_index.topk(title_embeds, k=k)
   final_candidates = []
-  final_candidates.append(list(image_candidates[0]))
+  final_candidates.append(list(candidates[0]))
   #final_candidates.append(list(title_candidates[0]))
   final_candidates = sum(final_candidates,[])
-  final_candidates = list(set(final_candidates))
+  # final_candidates = list(set(final_candidates))
   results_dict = {"image":[], "title":[]}
   for candidate in final_candidates:
     results_dict['image'].append(dataset['image'][candidate])