Merge pull request #10 from noisebridge/interaction-matrix-of-user

Interaction matrix of user

.gitignore

@@ -1,5 +1,12 @@
+interaction-matrix-of-user
+mv_0000001.txt
+mv_0000002.txt
+movie_titles.txt
+myenv/
+
 data
 out
 .env
 .pytest_cache
-__pycache__
+__pycache__
+ main

Pipfile

@@ -8,6 +8,7 @@ python-dotenv = "~=1.0"
 requests = "~=2.26"
 tqdm = "~=4.66"
 typer = "~=0.12"
+scipy = "*"
 
 [dev-packages]
 pytest = "~=8.3"
@@ -19,4 +20,4 @@ python_version = "3.12"
 
 [scripts]
 dev = "pipenv run python -m mediabridge.main"
-test = "pipenv run pytest"
+test = "pipenv run pytest"

lightfm_recommendation.py

@@ -0,0 +1,29 @@
+import numpy as np
+from lightfm import LightFM
+from lightfm.evaluation import auc_score, precision_at_k
+from scipy.sparse import coo_matrix
+
+# Example interaction data
+user_ids = [0, 0, 1, 1, 2]
+movie_ids = [0, 1, 1, 2, 2]
+ratings = [5, 4, 5, 4, 5]  # Only 4s and 5s for this example
+
+# Create sparse interaction matrix
+interaction_matrix = coo_matrix((ratings, (user_ids, movie_ids)), shape=(3, 3))
+
+# Initialize LightFM model with WARP loss
+model = LightFM(loss="warp")
+
+# Train the model
+model.fit(interaction_matrix, epochs=30, num_threads=2)
+
+# Predict scores for User 0
+scores = model.predict(0, np.arange(3))
+top_items = np.argsort(-scores)
+print("Top recommended items for User 0:", top_items)
+
+
+precision = precision_at_k(model, interaction_matrix, k=5).mean()
+auc = auc_score(model, interaction_matrix).mean()
+
+print(f"Precision at k=5: {precision}, AUC Score: {auc}")

mediabridge/data_processing/interaction_matrix.py

@@ -0,0 +1,71 @@
+import os
+import pickle
+
+import numpy as np
+from scipy.sparse import coo_matrix
+
+
+def list_rating_files(directory_path):
+    """List of files in the directory that start with mv_."""
+    for f in os.listdir(directory_path):
+        if f.startswith("mv_"):
+            yield os.path.join(directory_path, f)
+
+
+def create_interaction_matrix(directory_path, num_users, num_movies):
+    interaction_matrix = coo_matrix((num_users, num_movies), dtype=np.int8)
+    user_mapper = {}
+    current_user_index = 0
+
+    for file_path in list_rating_files(directory_path):
+        with open(file_path, "r") as file:
+            movie_id = int(file.readline().strip().replace(":", ""))
+            movie_idx = movie_id - 1
+
+            for line in file:
+                user_id, rating, _ = line.strip().split(",")
+                user_id = int(user_id)
+                rating = int(rating)
+
+                if rating < 4:
+                    continue
+
+                if user_id not in user_mapper:
+                    user_mapper[user_id] = current_user_index
+                    current_user_index += 1
+
+                user_idx = user_mapper[user_id]
+                interaction_matrix[user_idx, movie_idx] = rating
+
+    return interaction_matrix
+
+
+def save_matrix(matrix, output_file):
+    with open(output_file, "wb") as f:
+        pickle.dump(matrix, f)
+    print(f"Interaction matrix saved to {output_file}")
+
+
+def main():
+    """Main entry point to create and save the interaction matrix."""
+
+    # Configurations
+    data_directory = os.path.join(os.path.dirname(__file__), "../../data/")
+    output_directory = os.path.join(data_directory, "../output/")
+    output_file = os.path.join(output_directory, "interaction_matrix.pkl")
+
+    # Number of users and movies
+    num_users = 480189
+    num_movies = 17770
+
+    # Process Data
+    interaction_matrix = create_interaction_matrix(
+        data_directory, num_users, num_movies
+    )
+
+    # Save Data
+    save_matrix(interaction_matrix, output_file)
+
+
+if __name__ == "__main__":
+    main()