Create recommendation_utils.py

src/recommendation_utils.py

@@ -0,0 +1,65 @@
+import numpy as np
+import pandas as pd
+import pickle
+from keras.models import load_model
+
+# Lade Modelle und Daten
+def load_nn_model(path):
+    return load_model(path)
+
+def load_svd_model(path):
+    with open(path, "rb") as f:
+        return pickle.load(f)
+
+def load_trainset(path):
+    with open(path, "rb") as f:
+        return pickle.load(f)
+
+def fold_in_new_user(model, trainset, user_ratings, reg=5):
+    n_factors = model.n_factors
+    A = np.zeros((n_factors, n_factors))
+    b = np.zeros(n_factors)
+    bias_numerator, bias_denominator = 0, 0
+    global_mean = trainset.global_mean
+
+    for movie_id, rating in user_ratings.items():
+        try:
+            movie_id = int(movie_id)
+            inner_iid = trainset.to_inner_iid(movie_id)
+            qi = model.qi[inner_iid]
+            bi = model.bi[inner_iid]
+            A += np.outer(qi, qi)
+            b += qi * (rating - global_mean - bi)
+            bias_numerator += (rating - global_mean - bi)
+            bias_denominator += 1
+        except ValueError:
+            continue
+
+    A += reg * np.eye(n_factors)
+    new_user_factors = np.linalg.solve(A, b) if np.linalg.det(A) != 0 else np.linalg.pinv(A) @ b
+    new_user_bias = bias_numerator / bias_denominator if bias_denominator > 0 else 0
+    return new_user_factors, new_user_bias
+
+def recommend_with_svd(model, trainset, ratings_df, user_ratings, top_n=10):
+    new_user_factors, new_user_bias = fold_in_new_user(model, trainset, user_ratings)
+    global_mean = trainset.global_mean
+    movie_predictions = []
+    for inner_iid in range(trainset.n_items):
+        movie_id = trainset.to_raw_iid(inner_iid)
+        if movie_id not in user_ratings:
+            pred = global_mean + new_user_bias + model.bi[inner_iid] + np.dot(new_user_factors, model.qi[inner_iid])
+            movie_predictions.append({'movieId': int(movie_id), 'rating': pred})
+    df = pd.DataFrame(movie_predictions).sort_values("rating", ascending=False).head(top_n)
+    return df
+
+def recommend_with_nn(user_ratings, model, available_movies, top_n=10):
+    user_id = max(user_ratings.keys(), default=0) + 100000  # Dummy user ID
+    user_vector = np.array([user_id] * len(available_movies))
+    movie_vector = np.array(available_movies)
+    predictions = model.predict([user_vector, movie_vector], verbose=0)
+    df = pd.DataFrame({
+        'movieId': available_movies,
+        'rating': predictions.flatten()
+    })
+    df = df[~df["movieId"].isin(user_ratings.keys())]
+    return df.sort_values("rating", ascending=False).head(top_n)