标签：DeepFM
背景：广告是当今互联网商业变现最重要的模式之一，通过根据用户兴趣实时地定向投送广告，可以提高广告的转化率，提高广告营收。
目标：通过所提供的广告数据建立ML模型，预估出用户点击广告后是否会有效激活广告（成为付费用户）。
-
FM算法能捕捉到低阶的feature pair信息，但无法捕捉到高阶的feature pair信息
-
神经网络可以捕捉到高阶的feature pair信息
-
将FM与神经网络进行共享输入，不需要额外的特征工程，同时捕捉低阶和高阶的feature pair信息
df_train = x_train.copy()
df_train['label'] = y_train
plt.figure(figsize = (24, 16))
sns.heatmap(df_train.corr().round(2), annot = True, cmap = 'Greens')
-
app_second_categories与positionType和sitesetID和advertiserID和appID和appPlatform具有强相关关系
-
advertiserID与appID与appPlatform相互具有强相关关系
-
positionType与sitesetID具有强负相关关系
train_data.label.value_counts()0 3656266
1 93262
Name: label, dtype: int64
- 便签比为40:1，数据类别严重不均衡，考虑到样本量较大，对现有训练数据采用欠采样。
from imblearn.under_sampling import NearMiss
nm = NearMiss(version = 3, random_state=100, n_jobs = -1)
x_train_resam, y_train_resam = nm.fit_sample(x_train, y_train)from sklearn.preprocessing import LabelEncoder
le = LabelEncoder()
x_train_resam_le = x_train_resam.copy()
for col in x_train_resam.columns:
x_train_resam_le[col] = le.fit_transform(x_train_resam[col])feature_dict = {}
total_feature = 0
for col in x_train_resam.columns:
unique_val = x_train_resam[col].unique()
feature_dict[col] = dict(zip(unique_val, range(total_feature, total_feature + len(unique_val))))
total_feature += len(unique_val)
feature_value = x_train_resam.copy()
feature_index = x_train_resam.copy()
for col in x_train_resam.columns:
feature_value[col] = 1
feature_index[col] = feature_index[col].map(feature_dict[col])import tensorflow as tf
import numpy as np
import pandas as pd
dfm_params = {
'use_fm': True,
'use_deep': True,
'embedding_size': 8,
'dropout_fm': [1.0, 1.0],
'deep_layers': [32, 32],
'dropout_deep': [0.5, 0.5, 0.5],
'deep_layer_activation': tf.nn.relu,
'opoch': 30,
'bach_size': 1024,
'learning_rate': 0.001,
'optimizer': 'adam',
'batch_norm_dacay': 0.995,
'l2_reg': 0.01,
'verbose': True,
'eval_metric': 'logloss',
'random_seed': 100
}
dfm_params['feature_size'] = total_feature
dfm_params['field_size'] = len(x_train_resam.columns)
with tf.device('/gpu:0'):
feat_index = tf.placeholder(tf.int32, shape = [None, None], name = 'feat_index')
feat_value = tf.placeholder(tf.float32, shape = [None, None], name = 'feat_value')
label = tf.placeholder(tf.float32, shape = [None, 1], name = 'label')
weights = dict()
##embeddings
weights['feature_embeddings'] = tf.Variable(
tf.random_normal(
[dfm_params['feature_size'],
dfm_params['embedding_size']], 0.0, 0.01), name = 'feature_embeddings')
weights['feature_bias'] = tf.Variable(
tf.random_normal(
[dfm_params['feature_size'], 1],
0.0, 0.01), name = 'feature_bias')###weights[''feature_bias]是一次项的权重
##deep layers
num_layer = len(dfm_params['deep_layers'])
input_size = dfm_params['field_size'] * dfm_params['embedding_size']
glorot = np.sqrt(2.0 / (input_size + dfm_params['deep_layers'][0]))
weights['layer_0'] = tf.Variable(
np.random.normal(loc = 0, scale = glorot, size = (
input_size, dfm_params['deep_layers'][0])), dtype = np.float32)
weights['bias_0'] = tf.Variable(
np.random.normal(
loc = 0, scale = glorot, size = (
1, dfm_params['deep_layers'][0])), dtype = np.float32)
for i in range(1, num_layer):
glorot = np.sqrt(2.0 / (dfm_params['deep_layers'][i - 1] + dfm_params['deep_layers'][i]))
weights['layer_%d'%(i)] = tf.Variable(
np.random.normal(
loc = 0, scale = glorot,
size = (dfm_params['deep_layers'][i - 1],
dfm_params['deep_layers'][i])), dtype = np.float32)
weights['bias_%d'%(i)] = tf.Variable(
np.random.normal(
loc = 0, scale = glorot, size = (
1, dfm_params['deep_layers'][i])), dtype = np.float32)
##concat layer
if dfm_params['use_fm'] and dfm_params['use_deep']:
input_size = dfm_params[
'field_size'] + dfm_params['embedding_size'] + dfm_params['deep_layers'][-1]
elif dfm_params['use_fm']:
input_size = dfm_params['field_size'] + dfm_params['embedding_size']
else:
input_size = dfm_params['deep_layers'][-1]
glorot = np.sqrt(2.0 / (input_size + 1))
weights['concat_projection'] = tf.Variable(
np.random.normal(
loc = 0, scale = glorot, size = [input_size, 1]), dtype = np.float32)
weights['concat_bias'] = tf.Variable(tf.constant(0.01), dtype = np.float32)with tf.device('/gpu:0'):
##embeddings
embeddings = tf.nn.embedding_lookup(weights['feature_embeddings'], feat_index)
reshaped_feat_value = tf.reshape(feat_value,
shape = [-1, dfm_params['field_size'], 1])
print(reshaped_feat_value)
embeddings = tf.multiply(embeddings, reshaped_feat_value)
##fm
fm_first_order = tf.nn.embedding_lookup(weights['feature_bias'], feat_index)
fm_first_order = tf.reduce_sum(tf.multiply(fm_first_order, reshaped_feat_value), 2)
summed_features_emb = tf.reduce_sum(embeddings, 1)
summed_features_emb_square = tf.square(summed_features_emb)
squared_features_emb = tf.square(embeddings)
squared_features_emb_sum = tf.reduce_sum(squared_features_emb, 1)
fm_second_order = 0.5 * tf.subtract(
summed_features_emb_square, squared_features_emb_sum)
##deep
y_deep = tf.reshape(
embeddings, shape = [-1, dfm_params[
'field_size'] * dfm_params['embedding_size']])
for i in range(len(dfm_params['deep_layers'])):
y_deep = tf.add(tf.matmul(y_deep,
weights['layer_%d'%(i)]), weights['bias_%d'%(i)])
y_deep = tf.nn.relu(y_deep)
##final layer
if dfm_params['use_fm'] and dfm_params['use_deep']:
concat_input = tf.concat([fm_first_order, fm_second_order, y_deep], axis = 1)
elif dfm_params['use_fm']:
concat_input = tf.concat([fm_first_order, fm_second_order], axis = 1)
else:
concat_input = y_deep
out = tf.nn.sigmoid(
tf.add(tf.matmul(concat_input, weights['concat_projection']),
weights['concat_bias']))
loss = tf.losses.log_loss(tf.reshape(label, (-1,1)), out)
optimizer = tf.train.AdamOptimizer(learning_rate = dfm_params['learning_rate'],
beta1 = 0.9, beta2 = 0.999,
epsilon = 1e-8).minimize(loss)output: logloss is 0.015