Update

app.py

@@ -2,45 +2,34 @@ import gradio as gr
 import pandas as pd
 from datasets import load_dataset
 
-# 1) Load movie metadata and ratings from GroupLens URLs
-movies = load_dataset(
-    "csv",
-    data_files="https://files.grouplens.org/datasets/movielens/ml-latest-small/movies.csv",
-    split="train",
-)
-ratings = load_dataset(
-    "csv",
-    data_files="https://files.grouplens.org/datasets/movielens/ml-latest-small/ratings.csv",
-    split="train",
-)
+# 1) Load MovieLens metadata (small split, ~1.5K movies)
+ds = load_dataset("bstds/movielens", "small", split="train")
 
 # 2) Convert to pandas
-movies_df = movies.to_pandas()
-ratings_df = ratings.to_pandas()
-
-# 3) Infer a release year per movie by taking the earliest rating timestamp
-#    (assuming users start rating soon after release)
-ratings_df["year"] = pd.to_datetime(ratings_df["timestamp"], unit="s").dt.year
-first_year = ratings_df.groupby("movieId")["year"].min().reset_index()
-
-# 4) Merge metadata + inferred year, drop duplicates
-metadata = (
-    movies_df.merge(first_year, on="movieId", how="inner")
-    .rename(columns={"year": "release_year"})
-    .drop_duplicates(subset=["movieId"])
+df = ds.to_pandas()
+
+# 3) Extract release year from title, e.g. "Movie Title (1999)" → 1999
+df["release_year"] = (
+    df["title"]
+      .str.extract(r"\((\d{4})\)")    # capture 4-digit year
+      .astype(pd.Int64Dtype())         # allow missing values
 )
 
+# 4) Deduplicate metadata
+metadata = df[["title", "genres", "release_year"]].drop_duplicates()
+
 def recommend_by_genre_year(genre: str, year: int, top_k: int = 5) -> str:
-    # filter by genre substring (case-insensitive) and release_year ≥ year
+    # filter genre (case-insensitive substring)
     mask_genre = metadata["genres"].str.lower().str.contains(genre.lower())
-    mask_year = metadata["release_year"] >= year
-    candidates = metadata[mask_genre & mask_year]
+    # filter release year ≥ input year (drop rows missing year)
+    mask_year  = metadata["release_year"].fillna(0) >= year
 
+    candidates = metadata[mask_genre & mask_year]
     if candidates.empty:
-        return f"No {genre.title()} movies found from {year} onward."
+        return f"No '{genre.title()}' movies found from {year} onward."
 
     picks = candidates.sample(n=min(top_k, len(candidates)))
-    return "\n".join(f"• {row.title} ({int(row.release_year)})"
+    return "\n".join(f"• {row.title} ({int(row.release_year) if pd.notna(row.release_year) else 'Year N/A'})"
                      for _, row in picks.iterrows())
 
 iface = gr.Interface(
@@ -53,8 +42,8 @@ iface = gr.Interface(
     outputs="text",
     title="🎬 Genre & Year-Based Movie Recommender",
     description="""
-Pulls MovieLens “ml-latest-small” metadata & ratings live from GroupLens  
-to filter by genre and release year (inferred). No local files needed.
+Loads MovieLens metadata (small split) from the Hub, extracts release years from titles,
+and filters by genre substring & year. No local files needed.
 """,
 )