Upload app (1).py

app (1).py

@@ -0,0 +1,204 @@
+# -*- coding: utf-8 -*-
+"""app.ipynb
+
+Automatically generated by Colab.
+
+Original file is located at
+    https://colab.research.google.com/drive/1vb2j78WT7l9XQiXUBvl1VAtELSAzOUZJ
+"""
+
+import os
+import random
+import numpy as np
+import pandas as pd
+import faiss
+import gradio as gr
+from sentence_transformers import SentenceTransformer
+from huggingface_hub import InferenceClient
+import spotipy
+from spotipy.oauth2 import SpotifyClientCredentials
+from difflib import SequenceMatcher
+
+# ---------- Load data ----------
+CLEAN_CSV_PATH = "df_combined_clean.csv"
+EMB_PATH       = "df_embed.npz"
+INDEX_PATH     = "hnsw.index"
+
+df_combined = pd.read_csv(CLEAN_CSV_PATH)
+emb_data = np.load(EMB_PATH)
+df_embeddings = emb_data["df_embeddings"].astype("float32")
+index = faiss.read_index(INDEX_PATH)
+
+# ---------- Secrets ----------
+HF_TOKEN = os.getenv("HF_TOKEN")
+SPOTIFY_CLIENT_ID = os.getenv("SPOTIPY_CLIENT_ID")
+SPOTIFY_CLIENT_SECRET = os.getenv("SPOTIPY_CLIENT_SECRET")
+
+# ---------- Models ----------
+query_embedder = SentenceTransformer("all-mpnet-base-v2")
+hf_client = InferenceClient(model="meta-llama/Llama-2-7b-chat-hf", token=HF_TOKEN)
+
+sp = None
+if SPOTIFY_CLIENT_ID and SPOTIFY_CLIENT_SECRET:
+    auth = SpotifyClientCredentials(client_id=SPOTIFY_CLIENT_ID, client_secret=SPOTIFY_CLIENT_SECRET)
+    sp = spotipy.Spotify(auth_manager=auth)
+
+# ---------- Helper functions ----------
+def encode_query(text):
+    return query_embedder.encode([text], convert_to_numpy=True).astype("float32")
+
+def expand_with_llama(query):
+    if not hf_client:
+        return query
+    prompt = f"""You are helping someone search a lyrics catalog.
+If the input looks like lyrics or a singer name, return artist and song titles that match.
+Otherwise, return a short list of lyric-style keywords related to the input sentence.
+
+Input: {query}
+Output:"""
+    response = hf_client.text_generation(prompt, max_new_tokens=96, temperature=0.2, repetition_penalty=1.05)
+    return query + " " + str(response).strip().replace("\n", " ")
+
+def distances_to_similarity_pct(dists):
+    if len(dists) == 0: return np.array([])
+    dmin, dmax = dists.min(), dists.max()
+    if dmax - dmin == 0: return np.ones_like(dists) * 100
+    sims = 100 * (1 - (dists - dmin) / (dmax - dmin))
+    return sims
+
+def label_vibes(sim):
+    if sim >= 90: return "dead-on"
+    elif sim >= 80: return "strong vibes"
+    elif sim >= 70: return "adjacent"
+    elif sim >= 60: return "stretch but related"
+    else: return "pretty random"
+
+def semantic_search(query, k=10, random_extra=0, use_llama=True):
+    if not query.strip():
+        return pd.DataFrame(columns=["artist","song","similarity_pct","vibes","is_random"])
+    q_text = expand_with_llama(query) if use_llama else query
+    q_vec = encode_query(q_text)
+    dists, idxs = index.search(q_vec, k)
+    sem_df = df_combined.iloc[idxs[0]].copy()
+    sem_df["similarity_pct"] = distances_to_similarity_pct(dists[0])
+    sem_df["vibes"] = sem_df["similarity_pct"].apply(label_vibes)
+    sem_df["is_random"] = False
+    rand_df = pd.DataFrame()
+    if random_extra > 0:
+        chosen = np.random.choice(len(df_combined), size=min(random_extra, len(df_combined)), replace=False)
+        rand_df = df_combined.iloc[chosen].copy()
+        rand_df["similarity_pct"] = np.nan
+        rand_df["vibes"] = "pure random"
+        rand_df["is_random"] = True
+    results = pd.concat([sem_df, rand_df], ignore_index=True)
+    return results
+
+def lookup_spotify_track_smart(artist, song):
+    if not sp: return None, None
+    q = f"track:{song} artist:{artist}"
+    try:
+        results = sp.search(q, type="track", limit=3)
+        if not results["tracks"]["items"]:
+            return None, None
+        best = max(results["tracks"]["items"],
+                   key=lambda t: SequenceMatcher(None, t["name"].lower(), song.lower()).ratio())
+        return best["external_urls"]["spotify"], best["album"]["images"][0]["url"]
+    except Exception:
+        return None, None
+
+def search_pipeline(query, k=10, random_extra=0, use_llama=True):
+    res = semantic_search(query, k, random_extra, use_llama)
+    if res.empty or not sp:
+        res["spotify_url"], res["album_image"] = None, None
+        return res
+    urls, imgs = [], []
+    for _, r in res.iterrows():
+        u, i = lookup_spotify_track_smart(r["artist"], r["song"])
+        urls.append(u); imgs.append(i)
+    res["spotify_url"], res["album_image"] = urls, imgs
+    return res
+
+# ---------- HTML builders ----------
+def make_bg_style_html(query=None):
+    base_top, base_bottom = "#1e293b", "#020617"
+    return f"<style>:root {{--hf-bg-top:{base_top};--hf-bg-bottom:{base_bottom};}}</style>"
+
+def results_to_lux_html(results, query):
+    if results is None or results.empty:
+        return "<div class='lux-empty'>🎧 Describe a vibe to start.</div>"
+    cards = []
+    for _, r in results.iterrows():
+        cover = f"<img src='{r['album_image']}' class='lux-cover'/>" if isinstance(r.get("album_image"), str) else "♪"
+        btn = f"<a href='{r['spotify_url']}' target='_blank' class='lux-btn'>▶ Play on Spotify</a>" if isinstance(r.get("spotify_url"), str) else ""
+        sim = f"{int(r['similarity_pct'])}%" if pd.notnull(r['similarity_pct']) else ""
+        vibe = r['vibes']
+        chip = "🎲 random pick" if r['is_random'] else ""
+        cards.append(
+            f"<div class='lux-card'>{cover}<div class='lux-meta'><h3>{r['song']}</h3><p>{r['artist']}</p><span>{vibe}</span> {sim} {chip}{btn}</div></div>"
+        )
+    return "<div id='lux-wrapper'>" + "".join(cards) + "</div>"
+
+def get_random_vibe():
+    topics = ["late-night drives","dog bloopers","breakups","sunset beaches","college nostalgia"]
+    perspectives = ["first-person","third-person","group","inner monologue"]
+    contexts = ["dreamy","chaotic","romantic","melancholic"]
+    return f"Lyrics about {random.choice(topics)}, told in {random.choice(perspectives)}, {random.choice(contexts)}."
+
+def clear_all():
+    return "", "<div class='lux-empty'>🎧 Describe a vibe to start.</div>", make_bg_style_html()
+
+def search_with_bg(query, k, random_extra):
+    res = search_pipeline(query, int(k), int(random_extra), use_llama=True)
+    return results_to_lux_html(res, query), make_bg_style_html(query)
+
+# ---------- CSS ----------
+app_css = """
+@import url('https://fonts.googleapis.com/css2?family=Inter:wght@400;500;600;800;900&display=swap');
+body,.gradio-container{
+  background:radial-gradient(circle at 50% 0%,var(--hf-bg-top,#1e293b),var(--hf-bg-bottom,#020617) 80%)!important;
+  color:#e5e7eb;font-family:'Inter',system-ui,-apple-system,BlinkMacSystemFont,"Segoe UI",sans-serif!important;
+}
+.lux-card{background:rgba(255,255,255,0.05);border-radius:1rem;padding:1rem;margin:0.5rem;}
+.lux-cover{width:100%;border-radius:1rem;}
+.lux-btn{display:inline-block;margin-top:0.5rem;padding:0.4rem 0.8rem;background:#1db954;border-radius:9999px;color:white;text-decoration:none;}
+.primary-btn{background:#1db954;color:white;border-radius:1rem;}
+.secondary-btn{background:#334155;color:white;border-radius:1rem;}
+"""
+
+# ---------- Gradio UI ----------
+with gr.Blocks(css=app_css, title="HarmoniFind") as demo:
+    bg_style = gr.HTML(make_bg_style_html())
+    gr.HTML("""
+    <div id="hf-shell">
+      <div id="lux-header">
+        <div class="lux-subline">HARMONIFIND • LYRICS-DRIVEN SEMANTIC SEARCH</div>
+        <h1>Describe Your Song.</h1>
+        <p>We search by what the lyrics <strong>mean</strong>, not just titles or genres.</p>
+      </div>
+    </div>
+    """)
+    with gr.Column():
+        with gr.Row(variant="compact"):
+            input_box = gr.Textbox(
+                placeholder="Lyrics about a carefree road trip with too many snack stops",
+                show_label=False,
+                lines=3,
+                scale=5,
+            )
+            with gr.Column(scale=2,min_width=160):
+                search_btn   = gr.Button("Search",elem_classes=["primary-btn"])
+                surprise_btn = gr.Button("🎲 Surprise me",elem_classes=["secondary-btn"])
+                clear_btn    = gr.Button("Clear",elem_classes=["secondary-btn"])
+        with gr.Accordion("Search settings",open=False):
+            with gr.Row():
+                k_slider    = gr.Slider(5,50,value=10,step=1,label="# semantic matches")
+                rand_slider = gr.Slider(0,10,value=2,step=1,label="# extra random tracks")
+    output_html = gr.HTML()
+
+    input_box.submit(search_with_bg,[input_box,k_slider,rand_slider],[output_html,bg_style])
+    search_btn.click(search_with_bg,[input_box,k_slider,rand_slider],[output_html,bg_style])
+    surprise_btn.click(get_random_vibe,outputs=input_box).then(search_with_bg,[input_box,k_slider,rand_slider],[output_html,bg_style])
+    clear_btn.click(clear_all,None,[input_box,output_html,bg_style])
+
+if __name__ == "__main__":
+    demo.launch(server_name="0.0.0.0", server_port=7860)