fea_dict[idx] = dataset_iris[0][idx]
return npd, fea_dict
data_all, fea_dict = generate_sample_dataset_and_feature_dict(dataset_iris)
from sklearn.model_selection import train_test_split
data_train, data_test = train_test_split(data_all)
print(len(data_train), len(data_test))
# create dataset instance from structured ndarray data
import recsys.dataset as dataset
dataset = imp.reload(dataset)
train_dataset = dataset.Dataset(data_train,
total_users=1,
total_items=len(fea_dict),
implicit_negative=False,
name='Train')
test_dataset = dataset.Dataset(data_test,
total_users=1,
total_items=len(fea_dict),
implicit_negative=False,
name='Test')
# create featurizer for mapping: (user, item) -> vector
import recsys.featurizers.plain_featurizer as plain_featurizer
plain_featurizer = imp.reload(plain_featurizer)
class Featurizer(plain_featurizer.PlainFeaturizer):
"""
import recsys.dataset as dataset
train_data = np.load('lastfm_train.npy')
test_data = np.load('lastfm_test.npy')
total_users = max(
set(list(train_data['user_id']) + list(test_data['user_id']))) + 1
total_items = max(
set(list(train_data['item_id']) + list(test_data['item_id']))) + 1
print(total_users, total_items)
train_data[:2], test_data[:2]
# datasets
import recsys.dataset as dataset
dataset = imp.reload(dataset)
train_dataset = dataset.Dataset(train_data,
total_users,
total_items,
sortby='ts',
name='Train')
test_dataset = dataset.Dataset(test_data,
total_users,
total_items,
sortby='ts',
name='Test')
# hyperparamerters
dim_item_embed = 50 # dimension of item embedding
max_seq_len = 100 # the maxium length of user's listen history
num_units = 32 # Number of units in the RNN model
total_iter = int(1e3) # iterations for training
batch_size = 100 # training batch size
eval_iter = 200 # iteration of evaluation