From b64b7e7a5c7a65e3e4c7bd1e82543d0a43bd6fbe Mon Sep 17 00:00:00 2001
From: Guilherme Kruger <krugergui@gmail.com>
Date: Sun, 29 Oct 2023 23:10:17 +0100
Subject: [PATCH] Saving embeddings as array for easier loading

---
 src/backend/api/matching_algorithm.py       | 8 +++++++-
 src/backend/api/tokenization_n_embedding.py | 3 ++-
 2 files changed, 9 insertions(+), 2 deletions(-)

diff --git a/src/backend/api/matching_algorithm.py b/src/backend/api/matching_algorithm.py
index 788186e..2216747 100644
--- a/src/backend/api/matching_algorithm.py
+++ b/src/backend/api/matching_algorithm.py
@@ -1,3 +1,4 @@
+import json
 from typing import List
 import torch
 from sklearn.metrics.pairwise import cosine_similarity
@@ -22,11 +23,16 @@ def get_free_text_match(
     float: a number from -1 to 1 indicating cosine similarity
 
     """
+    
+
     if (candidate_embeddings) == None or (job_embeddings) == None:
         return 0
 
-    if type(candidate_embeddings) == str or type(job_embeddings) == str:
+    if candidate_embeddings == "" or type(job_embeddings) == "":
         return 0
+    
+    candidate_embeddings = json.loads(candidate_embeddings)
+    job_embeddings = json.loads(job_embeddings)
 
     return cosine_similarity(candidate_embeddings, job_embeddings)[0][0]
 
diff --git a/src/backend/api/tokenization_n_embedding.py b/src/backend/api/tokenization_n_embedding.py
index 1de1941..8181488 100644
--- a/src/backend/api/tokenization_n_embedding.py
+++ b/src/backend/api/tokenization_n_embedding.py
@@ -1,3 +1,4 @@
+import json
 import torch
 from typing import List
 from transformers import AutoTokenizer, AutoModel
@@ -43,7 +44,7 @@ def generate_embeddings(text: str, model_name: str=MODEL_NAME) -> List[List]:
     with torch.no_grad():
         text_outputs = model(**text_tokens)
     text_embeddings = text_outputs.last_hidden_state.mean(dim=1)
-    return text_embeddings
+    return json.dumps(text_embeddings.tolist())
 
 if __name__=="__main__":
     s = ""