- Download datasets from https://pan.baidu.com/s/1xrH1Q_O4zGxMNUbwNITpyg. The data is stored as the tfrecord format, with feature columns defined as:
features={
'user_id': tf.FixedLenFeature([1], tf.int64),
'shop_list': tf.FixedLenFeature([100], tf.int64),
'cate_list': tf.FixedLenFeature([100], tf.int64),
'catelevel1_list': tf.FixedLenFeature([100], tf.int64),
'brand_list': tf.FixedLenFeature([100], tf.int64),
'item_list': tf.FixedLenFeature([100], tf.int64),
'item_features_list': tf.FixedLenFeature([34*100], tf.float32),
'time_list': tf.FixedLenFeature([100], tf.int64),
'action_list': tf.FixedLenFeature([100], tf.int64),
'item_name_hash_list': tf.FixedLenFeature([15*100], tf.int64),
'i2q_term_hash_list': tf.FixedLenFeature([15*100], tf.int64),
'pv_reach_time': tf.FixedLenFeature([1], tf.int64),
'click': tf.FixedLenFeature([1], tf.float32),
'qid': tf.FixedLenFeature([1], tf.int64),
'query_features': tf.FixedLenFeature([7], tf.float32),
'query_hash_list': tf.FixedLenFeature([15], tf.int64),
'l1_cateid1': tf.FixedLenFeature([1], tf.int64),
'l1_cateid2': tf.FixedLenFeature([1], tf.int64),
'l1_cateid3': tf.FixedLenFeature([1], tf.int64),
'matchtype': tf.FixedLenFeature([1], tf.int64),
'cateid': tf.FixedLenFeature([1], tf.int64),
'triggerid': tf.FixedLenFeature([1], tf.int64),
'score_features': tf.FixedLenFeature([2], tf.float32),
'age_class': tf.FixedLenFeature([1], tf.int64),
'baby_stage': tf.FixedLenFeature([1], tf.int64),
'career_type': tf.FixedLenFeature([1], tf.int64),
'xftrl_features': tf.FixedLenFeature([14], tf.float32)
}where click is the label, behavior features in the user feature field are extracted from cate_list, catelevel1_list, item_list, item_features_list, time_list, action_list, item_name_hash_list, i2q_term_hash_list. Other features in the user, item, context feature fields are extracted from the remaining feature columns.
- Suppose the download path is: ~/QueryQR_dataset, then run:
python train_tfrecord.py --buckets "~/QueryQR_dataset/" --checkpointDir "~/log/"