def evaluate_model(model, item_valid_input, valid_labels, num_items, topK):
"""
Evaluate the performance (Hit_Ratio, NDCG) of top-K recommendation
Return: hits score & NDCG score.
"""
hits, ndcgs = [], []
valid_id_input = to_categorical(item_valid_input, num_classes=num_items)
valid_id_input = np.expand_dims(valid_id_input, axis=1)
valid_item_feature_input = embedding_matrix[item_valid_input, :]
valid_item_feature_input = np.expand_dims(valid_item_feature_input, axis=1)
predictions = model.predict([valid_id_input, valid_item_feature_input],
batch_size=args.batch_size)
#predictions = model.predict(valid_item_feature_input, batch_size=args.batch_size)
#predictions = model.predict([np.array(user_valid_input), np.array(item_valid_input)], batch_size=args.batch_size, verbose=0)
topk_ind = predictions.argsort()[:, ::-1][:, :topK]
for item in zip(topk_ind, valid_labels):
hr = ProjectUtility.getHitRatio(item[0], item[1])
ndcg = ProjectUtility.getNDCG(item[0], item[1])
hits.append(hr)
ndcgs.append(ndcg)
hits.append(hr)
ndcgs.append(ndcg)
return (hits, ndcgs)
def evaluate_model(model, user_valid_input, item_valid_input, valid_labels, num_items, topK):
"""
Evaluate the performance (Hit_Ratio, NDCG) of top-K recommendation
Return: score of each batch.
"""
print('\n# Evaluate on test data')
slice_y_val = keras.utils.to_categorical(valid_labels, num_items)
evaluate_loss = model.evaluate([np.array(user_valid_input), np.array(item_valid_input)],np.array(slice_y_val),batch_size=args.batch_size)
hits, ndcgs = [],[]
predictions = model.predict([np.array(user_valid_input), np.array(item_valid_input)], batch_size=args.batch_size, verbose=0)
print(len(predictions))
topk_ind = predictions.argsort()[:,::-1][:,:topK]
for item in zip(topk_ind,valid_labels):
hr = ProjectUtility.getHitRatio(item[0], item[1])
ndcg = ProjectUtility.getNDCG(item[0], item[1])
hits.append(hr)
ndcgs.append(ndcg)
hits.append(hr)
ndcgs.append(ndcg)
return (hits, ndcgs, evaluate_loss)
def evaluate_model(model, user_valid_input, item_valid_input,
valid_item_feature_input, valid_labels, topK):
"""
Evaluate the performance (Hit_Ratio, NDCG) of top-K recommendation
Return: score of each test rating.
"""
hits, ndcgs = [], []
predictions = model.predict(
[user_valid_input, item_valid_input, valid_item_feature_input],
batch_size=args.batch_size,
verbose=0)
topk_ind = predictions.argsort()[:, ::-1][:, :topK]
for item in zip(topk_ind, valid_labels):
hr = ProjectUtility.getHitRatio(item[0], item[1])
ndcg = ProjectUtility.getNDCG(item[0], item[1])
hits.append(hr)
ndcgs.append(ndcg)
hits.append(hr)
ndcgs.append(ndcg)
return (hits, ndcgs)
def evaluate_model(model, item_valid_input, valid_labels,num_items, topK):
"""
Evaluate the performance (Hit_Ratio, NDCG) of top-K recommendation
Return: score of each test rating.
"""
hits, ndcgs = [],[]
target_oh = to_categorical(valid_labels, num_classes=num_items)
input_oh = to_categorical(item_valid_input, num_classes=num_items)
input_oh = np.expand_dims(input_oh, axis=1)
predictions = model.predict(input_oh, batch_size=args.batch_size)
#predictions = model.predict([np.array(user_valid_input), np.array(item_valid_input)], batch_size=args.batch_size, verbose=0)
topk_ind = predictions.argsort()[:,::-1][:,:topK]
for item in zip(topk_ind,valid_labels):
hr = ProjectUtility.getHitRatio(item[0], item[1])
ndcg = ProjectUtility.getNDCG(item[0], item[1])
hits.append(hr)
ndcgs.append(ndcg)
hits.append(hr)
ndcgs.append(ndcg)
return (hits, ndcgs)
parser.add_argument('--valid_mask_path', type=str, default='data/preprocessed_data/valid_mask.dat')
parser.add_argument('--valid_target_path', type=str, default='data/preprocessed_data/valid_target.dat')
parser.add_argument('--item_dic_path', type=str, default='data/preprocessed_data/item_dic.txt')
parser.add_argument('--visitor_dic_path', type=str, default='data/preprocessed_data/visitor_dic.txt')
parser.add_argument('--item_features_path', type=str, default='data/preprocessed_data/item_features.csv')
parser.add_argument('--checkpoint_path', type=str, default='checkpoint/')
parser.add_argument('--batch_size', type=str, default=256)
parser.add_argument('--slice_size', type=str, default=256)
parser.add_argument('--topK', type=str, default=50)
parser.add_argument('--epochs', type=str, default=100)
parser.add_argument('--latent_dim', type=str, default=128)
parser.add_argument('--learning_rate', type=str, default=0.1)
parser.add_argument('--patience', type=str, default=5)
args = parser.parse_args()
visitor_dic=ProjectUtility.readItemDic(args.base_path+args.visitor_dic_path)
num_users=len(visitor_dic)
item_dic=ProjectUtility.readItemDic(args.base_path+args.item_dic_path)
num_items=len(item_dic)
#read training input data
with open(args.base_path+args.train_input_path, 'rb') as f:
train_inputlist = pickle.load(f)
train_inputlist = np.array(train_inputlist)
user_train_input=train_inputlist[:,0]
item_train_input=train_inputlist[:,1]
#read training input mask data
with open(args.base_path+args.train_mask_path, 'rb') as f:
train_mask_inputlist = pickle.load(f)
type=str,
default='data/preprocessed_data/visitor_dic.txt')
parser.add_argument('--item_features_path',
type=str,
default='data/preprocessed_data/item_features.csv')
parser.add_argument('--checkpoint_path', type=str, default='checkpoint/')
parser.add_argument('--batch_size', type=str, default=256)
parser.add_argument('--slice_size', type=str, default=256)
parser.add_argument('--topK', type=str, default=50)
parser.add_argument('--epochs', type=str, default=100)
parser.add_argument('--latent_dim', type=str, default=128)
parser.add_argument('--learning_rate', type=str, default=0.1)
parser.add_argument('--patience', type=str, default=5)
args = parser.parse_args()
visitor_dic = ProjectUtility.readItemDic(args.base_path +
args.visitor_dic_path)
num_users = len(visitor_dic)
item_dic = ProjectUtility.readItemDic(args.base_path + args.item_dic_path)
num_items = len(item_dic)
#read training input data
with open(args.base_path + args.train_input_path, 'rb') as f:
train_inputlist = pickle.load(f)
train_inputlist = np.array(train_inputlist)
user_train_input = train_inputlist[:, 0]
item_train_input = train_inputlist[:, 1]
#read training input mask data
with open(args.base_path + args.train_mask_path, 'rb') as f:
train_mask_inputlist = pickle.load(f)