def main(ratings_components=100, features_components=100, print_scores=False):
#data_path = '../data/goodbooks-10k/'
data_path = '../../goodbooks-10k/'
book_features = get_book_features(get_book_dataframe(data_path))
reduced_item_features, _, _ = reduce_matrix(
book_features, n_components=features_components)
goodreads_path = data_path + 'ratings.csv'
amazon_path = data_path + 'ratings_amazon.csv'
spr = get_ratings(goodreads_path, amazon_path, min_amazon_items=6)
n_folds = 5
scores = np.zeros((n_folds, 2))
kf = ColumnwiseKFold(n_folds, random_seed=30)
for i, (X, (user_incides, item_indices)) in enumerate(kf.split(spr)):
_, _, rating_VT = reduce_matrix(X, n_components=ratings_components)
reduced_item_ratings = rating_VT.T
items = get_reduced_joint(reduced_item_ratings, reduced_item_features)
sim = (cosine_similarity(items) + 1) / 2
scores[i, :] = evaluate(spr, X, sim, user_incides, item_indices)
if print_scores:
print_evaluation(scores[i, 0], scores[i, 1])
scores = np.mean(scores, axis=0)
if print_scores:
print('{0:d}-Fold Scores:')
print_evaluation(scores[0], scores[1])
return scores
def main(ratings_components=300, features_components=300, print_scores=False):
np.random.seed(42)
tf.set_random_seed(1984)
data_path = '../data/goodbooks-10k/'
book_features = get_book_features(get_book_dataframe(data_path))
reduced_item_features, _, _ = reduce_matrix(book_features, n_components=features_components)
goodreads_path = '../data/goodbooks-10k/ratings.csv'
amazon_path = '../data/amazon/ratings_amazon.csv'
spr = get_ratings(goodreads_path, amazon_path, min_amazon_items=6)
n_folds = 5
scores = np.zeros((n_folds, 2))
kf = ColumnwiseKFold(n_folds, random_seed=30)
for i, (X, (user_indices, item_indices)) in enumerate(kf.split(spr)):
_, _, rating_VT = reduce_matrix(X, n_components=ratings_components)
reduced_item_ratings = rating_VT.T
items = get_reduced_joint(reduced_item_ratings, reduced_item_features)
tf.reset_default_graph()
encoder = BookEncoder(user_input_dim=10000, book_input_dim=items.shape[1], user_hidden=150, book_hidden=150)
with tf.Session() as sess:
encoder.initialize(sess)
encoder.train(sess, X, items)
scores[i, :] = encoder.test(sess, spr, X, items, user_indices, item_indices)
if print_scores:
print_evaluation(scores[i, 0], scores[i, 1])
scores = np.mean(scores, axis=0)
if print_scores:
print('{0:d}-Fold Scores:'.format(n_folds))
print_evaluation(scores[0], scores[1])
return scores
def main():
"""
Sample program to verify the code.
This method will load and join ratings
"""
# Set this to where you save and load all data
# data_path = '../data/goodbooks-10k/'
data_path = '../../goodbooks-10k/'
goodreads_path = data_path + 'ratings.csv'
amazon_path = data_path + 'ratings_amazon.csv'
ratings = get_ratings(goodreads_path, amazon_path)
book_features = get_book_features(get_book_dataframe(data_path))
joint = get_joint(ratings.T, book_features, 30, 30)
print(joint.shape)