<i>Copyright (c) Recommenders contributors.</i>

<i>Licensed under the MIT License.</i>

Neural Collaborative Filtering on MovieLens dataset.

Neural Collaborative Filtering (NCF) is a well known recommendation algorithm that generalizes the matrix factorization problem with multi-layer perceptron.

This notebook provides an example of how to utilize and evaluate NCF implementation in the recommenders. We use a smaller dataset in this example to run NCF efficiently with GPU acceleration on a Data Science Virtual Machine.

python

%load_ext autoreload
%autoreload 2

python

import sys
import pandas as pd
import tensorflow as tf
tf.get_logger().setLevel('ERROR') # only show error messages

from recommenders.utils.timer import Timer
from recommenders.models.ncf.ncf_singlenode import NCF
from recommenders.models.ncf.dataset import Dataset as NCFDataset
from recommenders.datasets import movielens
from recommenders.datasets.python_splitters import python_chrono_split
from recommenders.evaluation.python_evaluation import (
    map, ndcg_at_k, precision_at_k, recall_at_k
)
from recommenders.utils.notebook_utils import store_metadata

print("System version: {}".format(sys.version))
print("Pandas version: {}".format(pd.__version__))
print("Tensorflow version: {}".format(tf.__version__))

Set the default parameters.

python

# top k items to recommend
TOP_K = 10

# Select MovieLens data size: 100k, 1m, 10m, or 20m
MOVIELENS_DATA_SIZE = '100k'

# Model parameters
EPOCHS = 50
BATCH_SIZE = 256

SEED = 42

1. Download the MovieLens dataset

python

df = movielens.load_pandas_df(
    size=MOVIELENS_DATA_SIZE,
    header=["userID", "itemID", "rating", "timestamp"]
)

2. Split the data using the Spark chronological splitter provided in utilities

python

train, test = python_chrono_split(df, 0.75)

Filter out any users or items in the test set that do not appear in the training set.

python

test = test[test["userID"].isin(train["userID"].unique())]
test = test[test["itemID"].isin(train["itemID"].unique())]

Write datasets to csv files.

python

train_file = "./train.csv"
test_file = "./test.csv"
train.to_csv(train_file, index=False)
test.to_csv(test_file, index=False)

Generate an NCF dataset object from the data subsets.

python

data = NCFDataset(train_file=train_file, test_file=test_file, seed=SEED)

3. Train the NCF model on the training data, and get the top-k recommendations for our testing data

NCF accepts implicit feedback and generates prospensity of items to be recommended to users in the scale of 0 to 1. A recommended item list can then be generated based on the scores. Note that this quickstart notebook is using a smaller number of epochs to reduce time for training. As a consequence, the model performance will be slighlty deteriorated.

python

model = NCF (
    n_users=data.n_users, 
    n_items=data.n_items,
    model_type="NeuMF",
    n_factors=4,
    layer_sizes=[16,8,4],
    n_epochs=EPOCHS,
    batch_size=BATCH_SIZE,
    learning_rate=1e-3,
    verbose=10,
    seed=SEED
)

python

with Timer() as train_time:
    model.fit(data)

print("Took {} seconds for training.".format(train_time))

In the movie recommendation use case scenario, seen movies are not recommended to the users.

python

with Timer() as test_time:
    users, items, preds = [], [], []
    item = list(train.itemID.unique())
    for user in train.userID.unique():
        user = [user] * len(item) 
        users.extend(user)
        items.extend(item)
        preds.extend(list(model.predict(user, item, is_list=True)))

    all_predictions = pd.DataFrame(data={"userID": users, "itemID":items, "prediction":preds})

    merged = pd.merge(train, all_predictions, on=["userID", "itemID"], how="outer")
    all_predictions = merged[merged.rating.isnull()].drop('rating', axis=1)

print("Took {} seconds for prediction.".format(test_time))

4. Evaluate how well NCF performs

The ranking metrics are used for evaluation.

python

eval_map = map(test, all_predictions, col_prediction='prediction', k=TOP_K)
eval_ndcg = ndcg_at_k(test, all_predictions, col_prediction='prediction', k=TOP_K)
eval_precision = precision_at_k(test, all_predictions, col_prediction='prediction', k=TOP_K)
eval_recall = recall_at_k(test, all_predictions, col_prediction='prediction', k=TOP_K)

print("MAP:\t%f" % eval_map,
      "NDCG:\t%f" % eval_ndcg,
      "Precision@K:\t%f" % eval_precision,
      "Recall@K:\t%f" % eval_recall, sep='\n')

python

# Record results for tests - ignore this cell
store_metadata("map", eval_map)
store_metadata("ndcg", eval_ndcg)
store_metadata("precision", eval_precision)
store_metadata("recall", eval_recall)
store_metadata("train_time", train_time.interval)
store_metadata("test_time", test_time.interval)