update app to blocks

app.py

@@ -5,9 +5,15 @@ import pandas as pd
 from datasets import load_dataset
 from sentence_transformers import SentenceTransformer
 
+idx = 0
 
-def build_doc_frame(df, idx=0):
-    doc = df.iloc[0]
+dataset = load_dataset("tollefj/rettsavgjoerelser_100samples_embeddings")
+model = SentenceTransformer("NbAiLab/nb-sbert-base")
+df = dataset["train"].to_pandas()
+
+
+def build_doc_frame(df, idx):
+    doc = df.iloc[idx]
     # as df:
     doc_df = pd.DataFrame(doc).T
     # keep only sentences + embedding:
@@ -21,17 +27,18 @@ def get_doc_embeddings(doc):
     return np.array(doc.embedding.tolist(), dtype="float32")
 
 
-def faiss_search(doc_idx, query_str, K=5):
-    # doc_idx is a choice option of (idx, text)
-    idx = doc_idx[0] - 1
-    newdoc = build_doc_frame(df, idx=idx)
+def faiss_search(doc_url, query_str, K=5):
+    global idx
+    # find idx from url:
+    doc_idx = df[df.url == doc_url].index[0]
+    idx = int(doc_idx)
+    newdoc = build_doc_frame(df, idx)
     embeddings = get_doc_embeddings(newdoc)
 
     faiss.normalize_L2(embeddings)
     index = faiss.IndexFlatIP(768)
     index.add(embeddings)
 
-    query_str = "Skade mellom kjøretøy"
     target_emb = model.encode([query_str])
     target_emb = np.array([target_emb.reshape(-1)])
     faiss.normalize_L2(target_emb)
@@ -43,30 +50,43 @@ def faiss_search(doc_idx, query_str, K=5):
     pretty_results = []
     for idx, score in zip(I[0], D[0]):
         pretty_results.append((round(float(score), 3), newdoc.iloc[idx].sentences))
-    pretty_results_str = "\n".join([f"{score}\t{sent}" for score, sent in pretty_results])
+    pretty_results_str = "\n".join([f"Score: {score}\t\t{sent}" for score, sent in pretty_results])
     top_k_str = f"Top {K} results for: {query_str}"
-    underlines = "__" * 40
 
     # return str:
-    return f"{top_k_str}\n{pretty_results_str}\n{underlines}"
+    return f"{top_k_str}\n{pretty_results_str}"
 
 
-dataset = load_dataset("tollefj/rettsavgjoerelser_100samples_embeddings")
-model = SentenceTransformer("NbAiLab/nb-sbert-base")
-df = dataset["train"].to_pandas()
+# def DropdownSummary():
+#     next_opts = df.iloc[idx].summary.tolist()
+#     return gr.Dropdown.update(choices=next_opts, label="Velg fra oppsummeringene")
+
+
+dropdown_opts = [doc.url for idx, doc in df.iterrows()]
+
+with gr.Blocks() as demo:
+    gr.Label("Lovdata rettsavgjørelser - semantisk søk")
+
+    case_dropdown = gr.Dropdown(label="Velg en rettsavgjørelse", choices=dropdown_opts, default=dropdown_opts[0])
+    # when case_dropdown changes, update the summary dropdown:
+    # idx_label = gr.Label(f"Current index: {idx}")
+
+    query = gr.Textbox(
+        label="Søk etter setninger",
+        lines=1,
+        placeholder="Kollisjon mellom to kjøretøy.",
+    )
+    k_slider = gr.Slider(minimum=1, maximum=10, label="Number of matches", value=5)
+
+    search_btn = gr.Button("Search")
+
+    output = gr.Textbox(label="Results", lines=10)
+
+    # from the selected URL, find the index in the df:
+    search_btn.click(faiss_search, inputs=[case_dropdown, query, k_slider], outputs=[output])
+
+    # clear_btn.click(None, inputs=[None, None], outputs=None)
+    # search_btn.click(faiss_search, inputs=[None, None, None], outputs=["text"])
+    # search_btn.click(faiss_search, inputs=[idx, query, k_slider], outputs=["text"])
 
-dropdown_opts = [(idx + 1, f"\t{doc.summary[0][:60]}...") for idx, doc in df.iterrows()]
-
-iface = gr.Interface(
-    fn=faiss_search,
-    inputs=[
-        gr.Dropdown(label="Select a court case", choices=dropdown_opts),
-        gr.Textbox(lines=2, placeholder="Your query here..."),
-        gr.Slider(minimum=1, maximum=10, label="Number of matches", value=5),
-    ],
-    outputs="text",
-    title="Lovdata rettsavgjørelser - semantisk søk",
-    description="Velg en rettsak og søk for å hente ut lignende setninger i saken",
-)
-
-iface.launch()
+demo.launch()