Update README.md

README.md

@@ -14,7 +14,7 @@ The parameters passed in the url request are :
 
 https://api.archives-ouvertes.fr/search/UNIV-COTEDAZUR/?q=docType_s:ART&fq=abstract_s:[%22%22%20TO%20*]&fq=domain_s:*shs*&fq=submittedDateY_i:[2020%20TO%202023]&fl=doiId_s,uri_s,title_s,subTitle_s,authFullName_s,producedDate_s,journalTitle_s,journalPublisher_s,abstract_s,domain_s,openAccess_bool
 
-The embeddings column stores the embeddings of the "combined" column values converted in vectors with the sentence-transformers/all-MiniLM-L6-v2 embeddinsg model.
+The embeddings corpus stores the embeddings of the "combined" column values converted in vectors with the sentence-transformers/all-MiniLM-L6-v2 embeddings model.
 
 ## Metadata extraction
 
@@ -31,54 +31,51 @@ print(total_cout)
 
 # Loop over the records and get metadata
 step = 500
-appended_data = []
+df = []
 for i in range(1, int(total_count), int(step)):
   url = f"{url}&rows={step}&start={i}&wt=csv"
   df = pd.read_csv(url, encoding="utf-8")
-  appended_data.append(df)
-appended_data = pd.concat(appended_data)
+  df.append(df)
+df = pd.concat(appended_data)
 
 # dedup
-appended_data = appended_data.drop_duplicates(subset=['uri_s'])
+df = appended_data.drop_duplicates(subset=['uri_s'])
 
-appended_data.shape
+df.shape
 # returns 2405
-```
-
-## Add embeddings (CPU)
 
-### Solution 1 : with HF Inference API
+# New column of concatenated textuel data
+df["combined"] = df.title_s + ". " + df.subTitle_s + ". " +df.abstract_s
 
+# Save
+df.to_csv("hal_data.csv", index=False, encoding="utf-8")
 ```
-import requests
-import json
-from typing import Optional, List, Dict, Any
-
-HF_TOKEN = "<hf_token>"
-model_id = "sentence-transformers/all-MiniLM-L6-v2"
-
-embeddings_api_url = f"https://api-inference.huggingface.co/pipeline/feature-extraction/{model_id}"
-headers = {"Authorization": f"Bearer {HF_TOKEN}"}
 
-def embeddings_query(text:str) -> List:
-    response = requests.post(embeddings_api_url, headers=headers, json={"inputs": text, "options":{"wait_for_model":True}})
-    return response.json()
+## Create embeddings with sentence-transformers library (CPU)
 
-df = appended_data.replace(np.nan, '')
-df['embeddings'] = df.combined.apply(lambda x:embeddings_query(x.strip()))
 ```
 
 ### Solution 2 : with sentence-transformers library
 
-```
+
 from sentence_transformers import SentenceTransformer
+import pickle
 
 model_id = "sentence-transformers/all-MiniLM-L6-v2"
 embedder = SentenceTransformer(model_id)
 
-def embeddings_query(text:str) -> List:
-    return embedder.encode(text,convert_to_tensor=True)
+# Test on simple piece of text
+embedder.encode(["The Ecology of Fear and Climate"],convert_to_tensor=True)
+
+# Create corpus
+corpus_embeddings = embedder.encode(df.combined.to_list(), show_progress_bar=True)
+
+# Save
+with open('hal_embeddings.pkl', "wb") as fOut:
+  pickle.dump(corpus_embeddings, fOut)
 
-df['embeddings'] = df.combined.apply(lambda x:embeddings_query(x.strip().to_list()))
+# Reload
+file = open("hal_embeddings.pkl",'rb')
+corpus_embeddings = pickle.load(file)
 
 ```