tensorflow 实现 ESMM

input_fn.py

#-*- coding: UTF-8 -*-
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import tensorflow as tf


FixedLenFeatureColumns=["label", "user_id", "creative_id", "has_target", "terminal",
        "hour", "weekday","template_category",
        "day_user_show", "day_user_click", "city_code","network_type"]
StringVarLenFeatureColumns = ["keyword"]  #特征长度不固定
FloatFixedLenFeatureColumns = ['creative_history_ctr']
StringFixedLenFeatureColumns = ["keyword_attention"]
StringFeatureColumns = ["device_type", "device_model", "manufacturer"]

DayShowSegs = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 37, 38, 39, 41, 42, 44, 46, 47, 49, 51, 54, 56, 59, 61, 65, 68, 72, 76, 81, 86, 92, 100, 109, 120, 134, 153, 184, 243, 1195]
DayClickSegs = [1, 2, 3, 6, 23]


def build_model_columns():
    """Builds a set of wide and deep feature columns."""
    # Continuous variable columns
    # hours_per_week = tf.feature_column.numeric_column('hours_per_week')

    creative_id = tf.feature_column.categorical_column_with_hash_bucket(
        'creative_id', hash_bucket_size=200000, dtype=tf.int64)
    # To show an example of hashing:
    has_target = tf.feature_column.categorical_column_with_identity(
        'has_target', num_buckets=3)
    terminal = tf.feature_column.categorical_column_with_identity(
        'terminal', num_buckets=10)
    hour = tf.feature_column.categorical_column_with_identity(
        'hour', num_buckets=25)
    weekday = tf.feature_column.categorical_column_with_identity(
        'weekday', num_buckets=10)
    day_user_show = tf.feature_column.bucketized_column(
        tf.feature_column.numeric_column('day_user_show', dtype=tf.int32), boundaries=DayShowSegs)
    day_user_click = tf.feature_column.bucketized_column(
        tf.feature_column.numeric_column('day_user_click', dtype=tf.int32), boundaries=DayClickSegs)

    city_code = tf.feature_column.categorical_column_with_hash_bucket(
        'city_code', hash_bucket_size=2000, dtype=tf.int64)

    network_type = tf.feature_column.categorical_column_with_identity(
        'network_type', num_buckets=20, default_value=19)

    device_type = tf.feature_column.categorical_column_with_hash_bucket(   #androidphone这些
        'device_type', hash_bucket_size=500000, dtype=tf.string
    )
    device_model = tf.feature_column.categorical_column_with_hash_bucket(  #型号如iPhone10  vivo X9
        'device_model', hash_bucket_size=200000, dtype=tf.string
    )
    manufacturer = tf.feature_column.categorical_column_with_hash_bucket(  #手机品牌 vivo iphone等
        'manufacturer', hash_bucket_size=50000, dtype=tf.string
    )


    deep_columns = [
        tf.feature_column.embedding_column(creative_id, dimension=15,combiner='sum'),
        tf.feature_column.embedding_column(has_target, dimension=15,combiner='sum'),
        tf.feature_column.embedding_column(terminal, dimension=15, combiner='sum'),
        tf.feature_column.embedding_column(hour, dimension=15, combiner='sum'),
        tf.feature_column.embedding_column(weekday, dimension=15, combiner='sum'),
        tf.fe