feat: filter selection by author

app.py

@@ -3,56 +3,70 @@ from sentence_transformers import SentenceTransformer
 import datasets
 import faiss
 import torch
+from sentence_transformers.util import semantic_search
+import time
 
-st.sidebar.text_input("Type your quote here")
 
-dataset = datasets.load_dataset('A-Roucher/english_historical_quotes', download_mode="force_redownload")
+if "initialized" not in st.session_state:
+    dataset = datasets.load_dataset('A-Roucher/english_historical_quotes', download_mode="force_redownload")
+    st.session_state.dataset = datasets.Dataset.from_dict(dataset['train'][:100])
 
-dataset = datasets.Dataset.from_dict(dataset['train'][:100])
+    st.session_state.all_authors = list(set(st.session_state.dataset['author']))
+    model_name = "sentence-transformers/all-MiniLM-L6-v2" # BAAI/bge-small-en-v1.5" # "Cohere/Cohere-embed-english-light-v3.0" # "sentence-transformers/all-MiniLM-L6-v2"
+    st.session_state.encoder = SentenceTransformer(model_name)
+    st.session_state.embeddings = st.session_state.encoder.encode(
+        st.session_state.dataset["quote"],
+        batch_size=4,
+        show_progress_bar=True,
+        convert_to_numpy=True,
+        normalize_embeddings=True,
+    )
+    st.session_state.initialized=True
 
-model_name = "sentence-transformers/all-MiniLM-L6-v2" # BAAI/bge-small-en-v1.5" # "Cohere/Cohere-embed-english-light-v3.0" # "sentence-transformers/all-MiniLM-L6-v2"
-encoder = SentenceTransformer(model_name)
+dataset_embeddings_tensor = torch.Tensor(st.session_state.embeddings)
 
-embeddings = encoder.encode(
-    dataset["quote"],
-    batch_size=4,
-    show_progress_bar=True,
-    convert_to_numpy=True,
-    normalize_embeddings=True,
-)
+sentence = "Knowledge of history is power."
 
-# dataset_embeddings = datasets.Dataset.from_dict({"embeddings": embeddings})
-# dataset_embeddings.add_faiss_index(column="embeddings")
+def search(query, selected_authors):
+    start = time.time()
+    if len(query.strip()) == 0:
+        return ""
+    
+    query_embedding = st.session_state.encoder.encode([query])
+    sentence_embedding_tensor = torch.Tensor(query_embedding)
 
-# dataset_embeddings.save_faiss_index('embeddings', 'output/index_alone.faiss')
+    if len(selected_authors) == 0:
+        author_indexes = [i for i in range(len(st.session_state.dataset))]
+    else:
+        author_indexes = [i for i in range(len(st.session_state.dataset)) if st.session_state.dataset['author'][i] in selected_authors]
+    hits = semantic_search(sentence_embedding_tensor, dataset_embeddings_tensor[author_indexes, :], top_k=5)
 
-# import faiss
+    indices = [author_indexes[i['corpus_id']] for i in hits[0]]
 
-# index = faiss.read_index('index_alone.faiss')
+    if len(indices) == 0:
+        return ""
+    result = "\n\n"
+    for i in indices:
+        result += f"###### {st.session_state.dataset['author'][i]}\n> {st.session_state.dataset['quote'][i]}\n----\n"
+    delay = "%.3f" % (time.time() - start)
+    return f"_Computation time: **{delay} seconds**_{result}"
 
-sentence = "Knowledge of history is power."
-sentence_embedding = encoder.encode([sentence])
-# scores, samples = dataset_embeddings.search(
-#     sentence_embedding, k=10
-# )
-sentence_embedding_tensor = torch.Tensor(sentence_embedding)
-dataset_embeddings_tensor = torch.Tensor(embeddings)
-from sentence_transformers.util import semantic_search
 
-author_indexes = list(range(10))
-hits = semantic_search(sentence_embedding_tensor, dataset_embeddings_tensor[author_indexes, :], top_k=5)
-
-list_hits = [author_indexes[i['corpus_id']] for i in hits[0]]
-print(list_hits)
-print(dataset)
-st.write(dataset.select(list_hits))
-
-# sentence_embedding = model.encode([sentence])
-# scores, sample_indexes = QUOTES_INDEX.search(
-#     sentence_embedding, k=k
-# )
-# quotes = QUOTES_DATASET.iloc[sample_indexes[0]]
-# author_descriptions_df = get_authors_descriptions(quotes['author'].unique())
-# quotes = quotes.merge(author_descriptions_df, on='author')
-# quotes["scores"] = scores[0]
-# quotes = quotes.sort_values("scores", ascending=True) # lower is better
+st.markdown(
+    """
+    <style>
+        div[data-testid="column"]
+        {
+            align-self:flex-end;
+        }
+    </style>
+    """,unsafe_allow_html=True
+)
+col1, col2 = st.columns([8, 2])
+text_input = col1.text_input("Type your idea here:")
+submit_button = col2.button("_Search quotes!_")
+selected_authors = st.multiselect("(Optional) - Restrict search to these authors:", st.session_state.all_authors)
+
+if submit_button:
+    st.markdown(search(text_input, selected_authors))
+