cols = [c for c in df_train.columns if c not in ["id", "target"]]
cols = [c for c in cols if c not in config.IGNORE_COLS]
X_train = df_train[cols].values
X_test = df_test[cols].values
y_train = df_train["target"].values
ids_test = df_test["id"].values
cat_features_indices = [
i for i, c in enumerate(cols) if c in config.CATEGORICAL_COLS
]
return df_train, df_test, X_train, y_train, X_test, ids_test, cat_features_indices
if __name__ == '__main__':
log("start to load data...")
df_train, df_test, X_train, y_train, X_test, ids_test, cat_features_indices = _load_data(
)
# folds
log("split folds")
folds = list(
StratifiedKFold(n_splits=config.NUM_SPLITS,
shuffle=True,
random_state=config.RANDOM_SEED).split(
X_train, y_train))
fd = FeatureDictionary(df_train=df_train,
df_test=df_test,
numeric_cols=config.NUMERIC_COLS,
ignored_cols=config.IGNORE_COLS)
log("parse data...")
import pandas as pd
import numpy as np
from sklearn.model_selection import StratifiedKFold
from example.data_reader import DataParser
from example.data_reader import FeatureDictionary
from example import config
from example.log import log
from load_data import *
if __name__ == '__main__':
log("start to load data...")
df_train, df_test, X_train, y_train, X_test, ids_test, cat_features_indices = load_ctr_data()
# folds
log("split folds")
folds = list(StratifiedKFold(n_splits=config.NUM_SPLITS, shuffle=True,
random_state=config.RANDOM_SEED).split(X_train, y_train))
'''#3.Random Forest Classifier'''
seed = 43
from sklearn.ensemble import RandomForestClassifier
rf = RandomForestClassifier(random_state = seed, n_estimators = 100)
'''#6.Decision Tree Classifier'''
from sklearn.tree import DecisionTreeClassifier
def _init_graph(self):
self.graph = tf.Graph()
with self.graph.as_default():
tf.set_random_seed(self.random_seed)
self.feature_index = tf.placeholder(tf.int32,
shape=[None, None],
name="feature_index")
self.feature_value = tf.placeholder(tf.float32,
shape=[None, None],
name="feature_value")
self.label = tf.placeholder(tf.float32,
shape=[None, 1],
name="label")
self.weights = self._initialize_weights()
self.dropout_keep_deep = tf.placeholder(tf.float32,
shape=[None],
name="dropout_keep_deep")
self.train_phase = tf.placeholder(tf.bool, name="train_phase")
# 1. embedding layer
self.embeddings = tf.nn.embedding_lookup(
self.weights["embedding_tensor"], self.feature_index) #
feature_value = tf.reshape(self.feature_value,
shape=[-1, self.field_dim, 1])
self.embeddings = tf.multiply(self.embeddings,
feature_value) # M * F * K
# 2. deep network
self.y_deep = tf.reshape(
self.embeddings,
shape=[-1, self.field_dim * self.embedding_dim])
self.y_deep = tf.nn.dropout(self.y_deep, self.dropout_keep_deep[0])
for i, layer_wide in enumerate(self.dnn_wides):
self.y_deep = tf.add(
tf.matmul(self.y_deep, self.weights["layer_%d" % i]),
self.weights["bias_%d" % i])
if self.batch_norm:
self.y_deep = self.batch_norm_layer(
self.y_deep,
train_phase=self.train_phase,
scope_bn="bn_%d" % i)
self.y_deep = self.dnn_activation(self.y_deep)
self.y_deep = tf.nn.dropout(self.y_deep,
self.dropout_keep_deep[1 + i])
# 3. cross network
input_size = self.field_dim * self.embedding_dim
self.y_cross_i = tf.reshape(self.embeddings,
shape=[-1, 1, input_size])
self.y_cross = tf.reshape(self.embeddings, shape=[-1, input_size])
self.y_cross_0 = tf.reshape(self.embeddings,
shape=[-1, 1, input_size])
for i in range(len(self.cross_wides)):
x0T_x_x1 = tf.reshape(tf.matmul(self.y_cross_0,
self.y_cross_i,
transpose_a=True),
shape=[-1, input_size])
self.y_cross_i = tf.add(
tf.reshape(tf.matmul(x0T_x_x1,
self.weights["cross_layer_%d" % i]),
shape=[-1, 1, input_size]), self.y_cross_i)
self.y_cross_i = tf.add(self.y_cross_i,
self.weights["cross_bias_%d" % i])
self.y_cross = tf.concat([
self.y_cross,
tf.reshape(self.y_cross_i, shape=[-1, input_size])
],
axis=1)
# 4. concatenate y_deep and y_cross
log("concatenating y_deep and y_cross")
if self.use_deep and self.use_cross:
concat_input = tf.concat([self.y_cross, self.y_deep], axis=1)
self.out = tf.add(
tf.matmul(concat_input, self.weights["concat_projection"]),
self.weights["concat_bias"])
elif self.use_deep:
concat_input = self.y_deep
self.out = tf.add(
tf.matmul(concat_input, self.weights["concat_projection"]),
self.weights["concat_bias"])
elif self.use_cross:
concat_input = self.y_cross
self.out = tf.add(
tf.matmul(concat_input, self.weights["concat_projection"]),
self.weights["concat_bias"])
# 5. loss
log("form loss")
self.out = tf.nn.sigmoid(self.out)
self.loss = tf.losses.log_loss(self.label, self.out)
# 6.regularization
log("regularization")
if self.l2_reg > 0.0:
self.loss += tf.contrib.layers.l2_regularizer(self.l2_reg)(
self.weights["concat_projection"])
for i in range(len(self.dnn_wides)):
self.loss += tf.contrib.layers.l2_regularizer(self.l2_reg)(
self.weights["layer_%d" % i])
# 7. optimizer
log("choose optimizer")
self.optimizer = tf.train.AdamOptimizer(
learning_rate=self.learning_rate,
beta1=0.9,
beta2=0.999,
epsilon=1e-8).minimize(self.loss)
# 8. init
log("run init...")
self.saver = tf.train.Saver()
init = tf.global_variables_initializer()
self.sess = self._init_session()
self.sess.run(init)
# number of params
total_parameters = 0
for variable in self.weights.values():
shape = variable.get_shape()
variable_parameters = 1
for dim in shape:
variable_parameters *= dim.value
total_parameters += variable_parameters
if self.verbose > 0:
print("#params: %d" % total_parameters)
def _init_graph(self):
self.graph = tf.Graph()
with self.graph.as_default():
tf.set_random_seed(self.random_seed)
self.feature_index = tf.placeholder(tf.int32,
shape=[None, None],
name="feature_index")
self.feature_value = tf.placeholder(tf.float32,
shape=[None, None],
name="feature_value")
self.label = tf.placeholder(tf.float32,
shape=[None, 1],
name="label")
self.weights = self._initialize_weights()
self.dropout_keep_deep = tf.placeholder(tf.float32,
shape=[None],
name="dropout_keep_deep")
self.train_phase = tf.placeholder(tf.bool, name="train_phase")
# 1. embedding layer
self.embeddings = tf.nn.embedding_lookup(
self.weights["embedding_tensor"], self.feature_index) #
feature_value = tf.reshape(self.feature_value,
shape=[-1, self.field_dim, 1])
self.embeddings = tf.multiply(self.embeddings,
feature_value) # M * F * K
# 2. deep network
self.y_deep = tf.reshape(
self.embeddings,
shape=[-1, self.field_dim * self.embedding_dim])
#self.y_deep = tf.nn.dropout(self.y_deep, self.dropout_keep_deep[0])
for i, layer_wide in enumerate(self.dnn_wides):
print("in deep %s" % i)
print(self.y_deep.shape.as_list())
print(self.weights["layer_%d" % i].shape.as_list())
print(self.weights["bias_%d" % i].shape.as_list())
self.y_deep = tf.add(
tf.matmul(self.y_deep, self.weights["layer_%d" % i]),
self.weights["bias_%d" % i])
# =============================================================================
# if self.batch_norm:
# self.y_deep = self.batch_norm_layer(self.y_deep, train_phase=self.train_phase, scope_bn="bn_%d" % i)
# =============================================================================
self.y_deep = self.dnn_activation(self.y_deep)
self.y_deep = tf.nn.dropout(self.y_deep,
self.dropout_keep_deep[1 + i])
# 3. cross network
input_size = self.field_dim * self.embedding_dim
self.y_cross = tf.reshape(self.embeddings,
shape=[-1, 1, input_size])
self.y_cross_0 = tf.reshape(self.embeddings,
shape=[-1, 1, input_size])
for i in range(len(self.cross_wides)):
print("in cross %d" % i)
print(self.y_cross_0.shape.as_list())
x0T_x_x1 = tf.reshape(tf.matmul(self.y_cross_0,
self.y_cross,
transpose_a=True),
shape=[-1, input_size])
print(x0T_x_x1.shape.as_list())
print(self.weights["cross_layer_%d" % i].shape.as_list())
self.y_cross = tf.add(
tf.reshape(tf.matmul(x0T_x_x1,
self.weights["cross_layer_%d" % i]),
shape=[-1, 1, input_size]), self.y_cross)
self.y_cross = tf.add(self.y_cross,
self.weights["cross_bias_%d" % i])
print("+++", self.y_cross.shape.as_list())
self.y_cross = tf.reshape(self.y_cross,
shape=[-1, self.cross_wides[0]])
# 4. concatenate y_deep and y_cross
log("concatenating y_deep and y_cross")
concat_input = tf.concat([self.y_cross, self.y_deep], axis=1)
self.out = tf.add(
tf.matmul(concat_input, self.weights["concat_projection"]),
self.weights["concat_bias"])
# 5. loss
log("form loss")
self.out1 = tf.nn.sigmoid(self.out)
#self.loss = tf.losses.log_loss(self.label, self.out1)
self.losses = tf.nn.sigmoid_cross_entropy_with_logits(
labels=self.label, logits=self.out)
self.loss = tf.reduce_mean(self.losses)
self.loss_summary = tf.summary.scalar("loss", self.loss)
# 6.regularization
log("regularization")
if self.l2_reg > 0.0:
self.loss += tf.contrib.layers.l2_regularizer(self.l2_reg)(
self.weights["concat_projection"])
for i in range(len(self.dnn_wides)):
self.loss += tf.contrib.layers.l2_regularizer(self.l2_reg)(
self.weights["layer_%d" % i])
# 7. optimizer
log("choose optimizer")
self.optimizer = tf.train.AdamOptimizer(
learning_rate=self.learning_rate,
beta1=0.9,
beta2=0.999,
epsilon=1e-8).minimize(self.loss)
# 8. init
log("run init...")
self.saver = tf.train.Saver()
init = tf.global_variables_initializer()
self.sess = self._init_session()
self.writer = tf.summary.FileWriter("logs/a3", self.sess.graph)
self.sess.run(init)
self.saver = tf.train.Saver(max_to_keep=2)
# =============================================================================
# model_path = os.path.abspath('.') + "\model2\my-model1-99"
# print(model_path)
# model_dict = '/'.join(model_path.split('/')[:-1])
# ckpt = tf.train.get_checkpoint_state(model_dict)
# self.saver.restore(self.sess, model_path)
# =============================================================================
# number of params
total_parameters = 0
for variable in self.weights.values():
shape = variable.get_shape()
variable_parameters = 1
for dim in shape:
variable_parameters *= dim.value
total_parameters += variable_parameters
if self.verbose > 0:
print("#params: %d" % total_parameters)