def build_model(self, is_training=True):
columns = get_feature_columns()
item_columns = fc.input_layer(self.other_features, columns['item'])
user_columns = fc.input_layer(self.other_features, columns['user'])
# 查找同ID的用户聚类特征
self.user_findid = self.table1.lookup(
self.other_features['FEA_CtxUid'])
self.user_embedding = tf.gather(self.user_feature, self.user_findid)
# 用户四种关系分离并进行attention
user_split = tf.split(self.user_embedding, 4, 1)
rela = tf.stack(user_split, 1)
rela_conv = tf.layers.conv1d(rela, 1, 1)
coef = tf.expand_dims(user_columns, 1)
coef = tf.layers.conv1d(coef, 1, 1)
out = tf.multiply(coef, rela_conv)
coefs = tf.nn.softmax(tf.nn.tanh(out), 1)
res = tf.multiply(coefs, rela)
res = tf.reduce_sum(res, 1)
item = self.cus_nn(item_columns, None, [
EMBEDDING_NUM * 2, EMBEDDING_NUM], is_training)
norm_item = tf.sqrt(tf.reduce_sum(tf.square(item), 1, True))
item_emb = tf.truediv(item, norm_item)
# for column in sorted(params['user_columns'], key=lambda x: x.name):
# print(column.name)
# 产出item embedding
if args.mode == 'sample':
self.prediction = {
'vid': self.other_features['FEA_SrcItemId'],
'item': item_emb,
}
else:
user = self.cus_nn(user_columns, None, [
EMBEDDING_NUM * 4, EMBEDDING_NUM * 2, EMBEDDING_NUM], is_training)
# attenton embedding拼接后送入全连接层
user_columns_out = tf.concat([user_columns, res], axis=1)
regularizer = tf.contrib.layers.l2_regularizer(scale=0.1)
user = tf.layers.dense(
inputs=user_columns_out,
units=EMBEDDING_NUM,
kernel_initializer=tf.contrib.layers.xavier_initializer(),
activation=partial(tf.nn.leaky_relu, alpha=0.2),
use_bias=True,
kernel_regularizer=regularizer,
)
# 残差连接
# user = user + net
# user = tf.concat([res, user],axis=1)
# user = self.cus_nn(user, None, [EMBEDDING_NUM], is_training)
norm_user = tf.sqrt(tf.reduce_sum(tf.square(user), 1, True))
user_emb = tf.truediv(user, norm_user)
self.cos_sim_raw = tf.reduce_sum(
tf.multiply(user_emb, item_emb), 1, True)
self.prob = tf.nn.sigmoid(self.cos_sim_raw)
def test_bucketized_column():
sample = {
'price': [[5.], [16], [25], [36]],
'time': [[2.], [6], [8], [15]]
}
price_column = feature_column.numeric_column('price')
bucket_price = feature_column.bucketized_column(price_column,
[10, 20, 30, 40])
price_bucket_tensor = feature_column.input_layer(sample, [bucket_price])
time_column = feature_column.numeric_column('time')
bucket_time = feature_column.bucketized_column(time_column, [5, 10, 12])
time_bucket_tensor = feature_column.input_layer(sample, [bucket_time])
with tf.Session() as session:
print(session.run([price_bucket_tensor, time_bucket_tensor]))
def test_categorical_column_with_hash_bucket():
#源数据
color_data = {'color': [[2], [5], [-1], [0]]} # 4行样本 shape=[4,1]
builder = _LazyBuilder(color_data)
# categorical_column
color_column = feature_column.categorical_column_with_hash_bucket(
'color', 7, dtype=tf.int32)
# tensor
color_column_tensor = color_column._get_sparse_tensors(builder) #稀疏表示
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
# 通过indicator_column，将稀疏的转换成dense，也就是one-hot形式，只是multi-hot
color_column_identy = feature_column.indicator_column(color_column)
#input_layer连接数据源和声明的column生成新的tensor
color_dense_tensor = feature_column.input_layer(color_data,
[color_column_identy])
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run([color_dense_tensor]))
def test_crossed_column():
""" crossed column测试 """
#源数据
featrues = {
'price': [['A'], ['B'], ['C']], # 0,1,2
'color': [['R'], ['G'], ['B']] # 0,1,2
}
# categorical_column
price = feature_column.categorical_column_with_vocabulary_list(
'price', ['A', 'B', 'C', 'D'])
color = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'])
#crossed_column 产生稀疏表示
p_x_c = feature_column.crossed_column([price, color], 16)
# 稠密表示
p_x_c_identy = feature_column.indicator_column(p_x_c)
# crossed_column 连接 源数据
p_x_c_identy_dense_tensor = feature_column.input_layer(
featrues, [p_x_c_identy])
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([p_x_c_identy_dense_tensor]))
def build_mode_norm_test(features, mode, params):
# Build the hidden layers, sized according to the 'hidden_units' param.
is_training = mode == tf.estimator.ModeKeys.TRAIN
fea_net = fc.input_layer(features, params['feature_columns'])
fea_net = tf.layers.batch_normalization(fea_net, training=is_training)
#x1 = tf.layers.dense(fea_net, units=256, activation=None, use_bias=False)
#hidden1 = tf.nn.relu(tf.layers.batch_normalization(x1, training=is_training), name='hidden1')
#x2 = tf.layers.dense(hidden1, units=128, activation=None, use_bias=False)
#hidden2 = tf.nn.relu(tf.layers.batch_normalization(x2, training=is_training), name='hidden2')
#net = tf.layers.dense(hidden2, units=64, activation=tf.tanh, name='user_vector_layer')
hidden1 = tf.layers.dense(fea_net,
units=128,
activation=tf.nn.relu,
name='hidden1')
#hidden2 = tf.layers.dense(hidden1, units=128, activation=tf.nn.relu, name='hidden2')
net = tf.layers.dense(hidden1,
units=64,
activation=tf.nn.relu,
name='user_vector_layer')
return net
def build_mode_norm(features, mode, params):
# Build the hidden layers, sized according to the 'hidden_units' param.
use_batch_norm = params['use_batch_norm']
is_training = mode == tf.estimator.ModeKeys.TRAIN
net = fc.input_layer(features, params['feature_columns'])
if use_batch_norm:
net = tf.layers.batch_normalization(net, training=is_training)
for units in params['hidden_units']:
if use_batch_norm:
x = tf.layers.dense(net,
units=units,
activation=None,
use_bias=False)
net = tf.nn.relu(
tf.layers.batch_normalization(x, training=is_training))
else:
net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
if use_batch_norm:
x = tf.layers.dense(net,
units=params['last_hidden_units'],
activation=None,
use_bias=False)
net = tf.nn.elu(tf.layers.batch_normalization(x, training=is_training),
name='user_vector_layer')
else:
net = tf.layers.dense(net,
units=params['last_hidden_units'],
activation=tf.nn.relu,
name='user_vector_layer')
return net
def test_embedding():
tf.set_random_seed(1)
#源数据
color_data = {
'color': [['R', 'G'], ['G', 'A'], ['B', 'B'], ['A', 'A']]
} # 4行样本
builder = _LazyBuilder(color_data)
# categorical_column 要想转为 embedding 先将源数据的clomn表达为categorical_column 这里只是声明没有源数据
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
# tensor 数据源 将数据源表达成tensor
color_column_tensor = color_column._get_sparse_tensors(builder)
#获取embedding_column； 第一个参数是：categorical_column； 第二个参数是维度
color_embedding_column = feature_column.embedding_column(color_column,
4,
combiner='sum')
# 转化为tensor input_layer(数据源，column) 连接起数据源和embedding_column
color_embeding_dense_tensor = feature_column.input_layer(
color_data, [color_embedding_column])
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
print('embeding' + '_' * 40)
print(session.run([color_embeding_dense_tensor]))
def test_categorical_column_with_hash_bucket():
# 1. Input features
color_data = {'color': [[2], [5], [-1], [0]]}
builder = _LazyBuilder(color_data)
# 2. Feature columns (Sparse)
color_column = feature_column.categorical_column_with_hash_bucket(
'color', 7, dtype=tf.int32)
color_column_tensor = color_column._get_sparse_tensors(builder)
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
# 2. Feature columns (Dense)
# Convert the Categorical Column to Dense Column
color_column_identity = feature_column.indicator_column(color_column)
# 3. Feature tensor
color_dense_tensor = feature_column.input_layer(color_data,
[color_column_identity])
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run([color_dense_tensor]))
def my_model(features, labels, mode, params):
net = fc.input_layer(features, params['feature_columns'])
# Build the hidden layers, sized according to the 'hidden_units' param.
for units in params['hidden_units']:
net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
net = tf.layers.dropout(
net,
params['dropout_rate'],
training=(mode == tf.estimator.ModeKeys.TRAIN))
my_head = tf.contrib.estimator.binary_classification_head(thresholds=[0.5])
# Compute logits (1 per class).
logits = tf.layers.dense(net,
my_head.logits_dimension,
activation=None,
name="my_model_output_logits")
optimizer = tf.train.AdagradOptimizer(
learning_rate=params['learning_rate'])
def _train_op_fn(loss):
return optimizer.minimize(loss, global_step=tf.train.get_global_step())
return my_head.create_estimator_spec(features=features,
mode=mode,
labels=labels,
logits=logits,
train_op_fn=_train_op_fn)
def test_weighted_categorical_column():
# 1. Input features
color_data = {
'color': [['R'], ['G'], ['B'], ['A']],
'weight': [[1.0], [2.0], [4.0], [8.0]]
}
# 2. Feature columns (Sparse)
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
# 2. Feature columns (Sparse)
color_weight_categorical_column \
= feature_column.weighted_categorical_column(color_column, 'weight')
builder = _LazyBuilder(color_data)
id_tensor, weight = color_weight_categorical_column._get_sparse_tensors(
builder)
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('weighted categorical' + '-' * 40)
print(session.run([id_tensor]))
print('-' * 40)
print(session.run([weight]))
# 2. Feature columns (Dense)
weighted_column = feature_column.indicator_column(
color_weight_categorical_column)
# 3. Feature tensor
weighted_column_dense_tensor = feature_column.input_layer(
color_data, [weighted_column])
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run([weighted_column_dense_tensor]))
def test_shared_embedding_column_with_hash_bucket():
# 1. Input features
color_data = {
'range': [[2, 2], [5, 5], [0, -1], [0, 0]],
'id': [[2], [5], [-1], [0]]
}
builder = _LazyBuilder(color_data)
# 2. Feature columns (Sparse)
color_column = feature_column.categorical_column_with_hash_bucket(
'range', 7, dtype=tf.int32)
color_column_tensor = color_column._get_sparse_tensors(builder)
# 2. Feature columns (Sparse)
color_column2 = feature_column.categorical_column_with_hash_bucket(
'id', 7, dtype=tf.int32)
color_column_tensor2 = color_column2._get_sparse_tensors(builder)
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('not use input_layer' + '_' * 40)
print(session.run([color_column_tensor.id_tensor]))
print(session.run([color_column_tensor2.id_tensor]))
# 2. Feature columns (Dense)
color_column_embed = feature_column.shared_embedding_columns(
[color_column2, color_column], 3, combiner='sum')
print(type(color_column_embed))
# 3. Feature tensor
color_dense_tensor = feature_column.input_layer(color_data,
color_column_embed)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run(color_dense_tensor))
def test_embedding():
tf.set_random_seed(1)
# 1. Input features
color_data = {'color': [['R', 'G'], ['G', 'A'], ['B', 'B'], ['A', 'A']]}
builder = _LazyBuilder(color_data)
# 2. Feature columns (Sparse)
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
color_column_tensor = color_column._get_sparse_tensors(builder)
with tf.Session() as session:
#session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
# 2. Feature columns (Dense)
color_embedding = feature_column.embedding_column(color_column,
4,
combiner='sum')
# 3. Feature tensor
color_embedding_dense_tensor = feature_column.input_layer(
color_data, [color_embedding])
with tf.Session() as session:
# Embedding needs variables (weights) to do the embedding
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('embedding' + '_' * 40)
print(session.run([color_embedding_dense_tensor]))
def test_categorical_column_with_vocabulary_list():
color_data = {
'color': [['R', 'R'], ['G', 'R'], ['B', 'G'], ['A', 'A']]
} # 4行样本
builder = _LazyBuilder(color_data)
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
color_column_tensor = color_column._get_sparse_tensors(builder)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
# 将稀疏的转换成dense，也就是one-hot形式，只是multi-hot
color_column_identy = feature_column.indicator_column(color_column)
color_dense_tensor = feature_column.input_layer(color_data,
[color_column_identy])
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run([color_dense_tensor]))
def test_multi_value_embedding():
color_data = {
'color': [['G', 'G'], ['G', 'B'], ['B', 'B'], ['G', 'R'], ['R', 'R'],
['B', 'R']]
}
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
color_embeding = feature_column.embedding_column(color_column, 7)
color_embeding_dense_tensor = feature_column.input_layer(
color_data, [color_embeding])
builder = _LazyBuilder(color_data)
color_column_tensor = color_column._get_sparse_tensors(builder)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('embeding' + '-' * 40)
print(session.run([color_embeding_dense_tensor]))
def test_weighted_cate_column():
# !!! id=''代表missing，其对应的weight只能为0，否则会导致id和weight长度不一致而报错
# !!! 而且weight必须是float型，输入int会报错
x_values = {
'id': [[b'a', b'z', b'a', b'c'], [b'b', b'', b'd', b'b']],
'weight': [[1.0, 2.0, -3.0, 4.0], [5.0, 0.0, 7.0, -8.0]]
}
builder = _LazyBuilder(x_values) # lazy representation of input
# ================== define ops
sparse_id_featcol = feature_column.categorical_column_with_vocabulary_list(
'id', ['a', 'b', 'c'], dtype=tf.string, default_value=-1)
sparse_featcol = feature_column.weighted_categorical_column(
categorical_column=sparse_id_featcol, weight_feature_key='weight')
x_sparse_tensor = sparse_featcol._get_sparse_tensors(builder)
# indicator_column将sparse tensor转换成dense MHE格式, shape=[batch_size, #tokens]
# 其中的权重是这个token出现的所有权重的总和
dense_featcol = feature_column.indicator_column(sparse_featcol)
x_dense_tensor = feature_column.input_layer(x_values, [dense_featcol])
# ================== run
with tf.Session() as sess:
# 必须initialize table，否则报错
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
id_sparse_value, weight_sparse_value = sess.run(
[x_sparse_tensor.id_tensor, x_sparse_tensor.weight_tensor])
print("************************* sparse id tensor")
# sparse tensor's id_tensor保持与原始输入相同的形状，[batch_size, max_tokens_per_example]=[2,4]
# SparseTensorValue(indices=array(
# [[0, 0],
# [0, 1],
# [0, 2],
# [0, 3],
# [1, 0],
# [1, 2],
# [1, 3]]), values=array([ 0, -1, 0, 2, 1, -1, 1]), dense_shape=array([2, 4]))
print(id_sparse_value)
print("************************* sparse weight tensor")
# sparse tensor's weight_tensor保持与原始输入相同的形状，[batch_size, max_tokens_per_example]=[2,4]
# SparseTensorValue(indices=array(
# [[0, 0],
# [0, 1],
# [0, 2],
# [0, 3],
# [1, 0],
# [1, 2],
# [1, 3]]), values=array([ 1., 2., -3., 4., 5., 7., -8.], dtype=float32), dense_shape=array([2, 4]))
print(weight_sparse_value)
print("************************* dense MHE tensor")
# indicator_column将sparse tensor按照MHE的方式转化成dense tensor，shape=[batch_size, total_tokens_in_vocab]
# 其中的每个数值是该token出现的所有权重的总和
# [[-2. 0. 4.]
# [ 0. -3. 0.]]
print(sess.run(x_dense_tensor))
def test_shared_embedding_column_with_hash_bucket():
color_data = {
'color': [[2, 2], [5, 5], [0, -1], [0, 0]],
'color2': [[2], [5], [-1], [0]]
} # 4行样本
builder = _LazyBuilder(color_data)
color_column = feature_column.categorical_column_with_hash_bucket(
'color', 7, dtype=tf.int32)
color_column_tensor = color_column._get_sparse_tensors(builder)
color_column2 = feature_column.categorical_column_with_hash_bucket(
'color2', 7, dtype=tf.int32)
color_column_tensor2 = color_column2._get_sparse_tensors(builder)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('not use input_layer' + '_' * 40)
print(session.run([color_column_tensor.id_tensor]))
print(session.run([color_column_tensor2.id_tensor]))
# 将稀疏的转换成dense，也就是one-hot形式，只是multi-hot
color_column_embed = feature_column.shared_embedding_columns(
[color_column2, color_column], 3, combiner='sum')
print(type(color_column_embed))
color_dense_tensor = feature_column.input_layer(color_data,
color_column_embed)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('use input_layer' + '_' * 40)
print(session.run(color_dense_tensor))
def pratise():
d = {'x': [[32], [16], [38], [98]]}
cd = feature_column.numeric_column('x')
bcd = feature_column.bucketized_column(cd, [10, 20, 40, 60])
fcd = feature_column.input_layer(d, [bcd])
with tf.Session() as sess:
print(sess.run(fcd))
def test_bucketized_column():
price = {'price': [[5.], [15.], [25.], [35.]]} # 4行样本 shape =[4,1]
price_column = feature_column.numeric_column('price')
bucket_price = feature_column.bucketized_column(price_column,
[10, 20, 30, 40])
price_bucket_tensor = feature_column.input_layer(price, [bucket_price])
with tf.Session() as session:
print(session.run([price_bucket_tensor]))
def test_weighted_categorical_feature_embedding():
color_data = {
'color': [['R', 'R'], ['G', 'G'], ['B', 'B'], ['G', 'R'], ['G', 'B'],
['B', 'R']],
'weight': [[0.5, 0.5], [0.5, 0.5], [0.5, 0.5], [0.3, 0.2], [0.4, 0.3],
[0.4, 0.6]]
} # 6行样本
color_column = feature_column.categorical_column_with_vocabulary_list(
'color', ['R', 'G', 'B'], dtype=tf.string, default_value=-1)
color_embeding = feature_column.embedding_column(color_column,
7,
combiner="sum")
color_embeding_dense_tensor = feature_column.input_layer(
color_data, [color_embeding])
color_weight_categorical_column = feature_column.weighted_categorical_column(
color_column, 'weight')
color_embeding_weighted = feature_column.embedding_column(
color_weight_categorical_column, 7, combiner="sum")
color_embeding_dense_tensor_2 = feature_column.input_layer(
color_data, [color_embeding_weighted])
builder = _LazyBuilder(color_data)
color_column_tensor = color_column._get_sparse_tensors(builder)
color_weighted_tensor = color_weight_categorical_column._get_sparse_tensors(
builder) ## is a pair (id_tensor, weight_tensor)
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print(session.run([color_column_tensor.id_tensor]))
print("color column weight:")
print(color_column_tensor.weight_tensor)
print("color column weighted categorical, weight:")
print(session.run([color_weighted_tensor.id_tensor]))
print(session.run([color_weighted_tensor.weight_tensor]))
with tf.Session() as session:
session.run(tf.global_variables_initializer())
session.run(tf.tables_initializer())
print('embeding' + '-' * 40)
print(session.run([color_embeding_dense_tensor]))
print('embeding weighted categorical column')
print(session.run([color_embeding_dense_tensor_2]))
def build_item_model(features, mode, params):
with tf.variable_scope("item_side",
partitioner=tf.fixed_size_partitioner(len(
FLAGS.ps_hosts.split(",")),
axis=0)):
item_uuid_embed = fc.input_layer(
features, params["feature_configs"].all_columns["itemID"])
item_dense = tf.nn.l2_normalize(item_uuid_embed)
return item_dense
def build_mode(features, mode, params, columns):
# import pdb;pdb.set_trace()
net = fc.input_layer(features, columns)
# Build the hidden layers, sized according to the 'hidden_units' param.
for units in params['hidden_units']:
net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
net = tf.layers.dropout(net, params['dropout_rate'], training=(mode == tf.estimator.ModeKeys.TRAIN))
return net
def dupn_model_fn(features, labels, mode, params):
behvr_emb, property_emb, item_emb = get_behavior_embedding(
params, features)
print("behvr_emb shape:", behvr_emb.shape)
print("property_emb shape:", property_emb.shape)
print("item_emb shape:", item_emb.shape)
inputs = tf.concat([behvr_emb, property_emb], -1)
print("lstm inputs shape:", inputs.shape)
lstm_cell = tf.nn.rnn_cell.BasicLSTMCell(num_units=params["num_units"])
#initial_state = lstm_cell.zero_state(params["batch_size"], tf.float32)
outputs, state = tf.nn.dynamic_rnn(lstm_cell, inputs, dtype=tf.float32)
print("lstm output shape:", outputs.shape)
masks = tf.cast(features["behaviorPids"] >= 0, tf.float32)
user = fc.input_layer(features, params["user_feature_columns"])
context = tf.concat([user, item_emb], -1)
print("attention context shape:", context.shape)
sequence = attention(outputs, context, params, masks)
print("sequence embedding shape:", sequence.shape)
other = fc.input_layer(features, params["other_feature_columns"])
net = tf.concat([sequence, item_emb, other], -1)
# Build the hidden layers, sized according to the 'hidden_units' param.
for units in params['hidden_units']:
net = tf.layers.dense(net, units=int(units), activation=tf.nn.relu)
if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
net = tf.layers.dropout(
net,
params['dropout_rate'],
training=(mode == tf.estimator.ModeKeys.TRAIN))
# Compute logits
logits = tf.layers.dense(net, 1, activation=None)
optimizer = optimizers.get_optimizer_instance(params["optimizer"],
params["learning_rate"])
my_head = tf.contrib.estimator.binary_classification_head(thresholds=[0.5])
return my_head.create_estimator_spec(
features=features,
mode=mode,
labels=labels,
logits=logits,
train_op_fn=lambda loss: optimizer.minimize(
loss, global_step=tf.train.get_global_step()))
def build_mode(features, mode, params):
net = fc.input_layer(features, params['feature_columns'])
# Build the hidden layers, sized according to the 'hidden_units' param.
for units in params['hidden_units']:
net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
net = tf.layers.dropout(net, params['dropout_rate'], training=(mode == tf.estimator.ModeKeys.TRAIN))
# Compute logits
logits = tf.layers.dense(net, 1, activation=None)
return logits
def practise():
fx = {'x': [['a', 'a'], ['b', 'c'], ['c', 'e'], ['d', ''], ['e', 'f']]}
fc = feature_column.categorical_column_with_hash_bucket('x', 5)
fic = feature_column.indicator_column(fc)
t2 = fc._get_sparse_tensors(_LazyBuilder(fx)).id_tensor
tsor = feature_column.input_layer(fx, fic)
with tf.Session() as sess:
print(sess.run(t2))
print(sess.run(tsor))
def test_identity_feature_column():
sample = {'price': [[1], [2], [3], [0]]}
# price_column = feature_column.numeric_column('price')
price_column = feature_column.categorical_column_with_identity(
key='price', num_buckets=4)
indicator = feature_column.indicator_column(price_column)
price_column_tensor = feature_column.input_layer(sample, [indicator])
with tf.Session() as session:
print(session.run([price_column_tensor]))
def inference(self, feats):
# embedding_columns, order_columns, spacetime_columns, user_columns = make_columns()
embedding_columns, order_columns, spacetime_columns, user_columns = make_columns_with_normalizer()
# TODO: feats 按columns拆分
with tf.name_scope('embedding_columns'):
embedding_tensor = fc.input_layer(feats, embedding_columns)
with tf.name_scope('order_columns'):
order_tensor = fc.input_layer(feats, order_columns)
with tf.name_scope('spacetime_columns'):
spacetime_tensor = fc.input_layer(feats, spacetime_columns)
input_tensor = tf.concat([embedding_tensor, order_tensor, spacetime_tensor], axis=1, name='input_concat')
eta_d = tf.concat([net('eta_d_%d'%ix, input_tensor, self.layer_units) for ix in range(self.d_k)], axis=1)
eta_c = tf.concat([net('eta_c_%d'%ix, input_tensor, self.layer_units) for ix in range(self.c_k)], axis=1)
d_softmax_logits = net('logits_d', input_tensor, self.layer_units, self.d_k)
c_softmax_logits = net('logits_c', input_tensor, self.layer_units, self.c_k)
return eta_d, eta_c, d_softmax_logits, c_softmax_logits
def test_bucketized_column():
# 1. Input features
price = {'price': [[15.], [5.], [35.], [25.]]}
# 2. Feature columns (Dense)
price_column = feature_column.numeric_column('price')
# 2. Feature columns (Dense): bucketized_column is both Dense and
# Categorical
bucket_price = feature_column.bucketized_column(price_column, [10, 20, 30])
# 3. Feature tensor
price_bucket_tensor = feature_column.input_layer(price, [bucket_price])
with tf.Session() as session:
print(session.run([price_bucket_tensor]))
def test_cate_featcol_with_vocablist():
# ================== prepare input
# 1. 为什么要用b前缀，这是因为通过input_fn读进来的字符串都有b前缀
# 而我要测试的是，如果我传入的vocab list是普通的str，能否与这些b匹配成功
# 2. '' represents missing, feature_column treats them "ignored in sparse tensor, and 0 in dense tensor"
# 3. 'z' represents OOV, feature_column treats them "-1 in sparse tensor, and 0 in dense tensor"
# 4. duplicates should be merged in dense tensor by summing up their occurrence
x_values = {'x': [[b'a', b'z', b'a', b'c'], [b'b', b'', b'd', b'b']]}
builder = _LazyBuilder(x_values) # lazy representation of input
# ================== define ops
sparse_featcol = feature_column.categorical_column_with_vocabulary_list(
'x', ['a', 'b', 'c'], dtype=tf.string, default_value=-1)
x_sparse_tensor = sparse_featcol._get_sparse_tensors(builder)
# 尽管一行中有重复，但是并没有合并，所以压根就没有weight
# 只是导致id_tensor中会出现重复数值而已，而导致embedding_lookup_sparse时出现累加
assert x_sparse_tensor.weight_tensor is None
# indicator_column将sparse tensor转换成dense MHE格式，注意第一行有重复，所以结果应该是multi-hot
dense_featcol = feature_column.indicator_column(sparse_featcol)
x_dense_tensor = feature_column.input_layer(x_values, [dense_featcol])
# ================== run
with tf.Session() as sess:
# 必须initialize table，否则报错
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
print("************************* sparse tensor")
# 结果证明：
# 1. 输入数据用b前缀的字符串，能够匹配上vocab list中的str
# 2. 注意第二行只有两个元素，''在sparse tensor中被忽略掉了
# 3. 'z','d'代表oov，sparse tensor中被映射成-1
# 4. sparse tensor的dense_shape与原始输入的shape相同
# [SparseTensorValue(indices=array([[0, 0],
# [0, 1],
# [0, 2],
# [0, 3],
# [1, 0],
# [1, 2],
# [1, 3]]), values=array([0, -1, 0, 2, 1, -1, 1]), dense_shape=array([2, 4]))]
print(sess.run([x_sparse_tensor.id_tensor]))
print("************************* dense MHE tensor")
# 结果证明：
# 1. 在dense表示中，duplicates的出现次数被加和，使用MHE
# 2. 无论是原始的missing（或许是由padding造成的），还是oov，在dense结果中都不出现
# 3. dense_tensor的shape=[#batch_size, vocab_size]
# [[2. 0. 1.]
# [0. 2. 0.]]
print(sess.run(x_dense_tensor))
def build_model_net(features, mode, params):
net = fc.input_layer(features, params['feature_columns'])
# net = tf.layers.batch_normalization(net, training=(mode == tf.estimator.ModeKeys.TRAIN))
# Build the hidden layers, sized according to the 'hidden_units' param.
for units in params['hidden_units']:
net = tf.layers.dense(net, units=units, activation=tf.nn.relu)
if 'dropout_rate' in params and params['dropout_rate'] > 0.0:
net = tf.layers.dropout(
net,
params['dropout_rate'],
training=(mode == tf.estimator.ModeKeys.TRAIN))
print("net node count", net.shape[-1].value)
logits = tf.layers.dense(net, units=1)
return logits
def emb():
xb = {'x': [['a', 'b'], ['a', 'c'], ['b', 'c']]}
x = {
'x': [['a', 'b'], ['b', 'c'], ['c', ''], ['', '']]
} #可以是多值变量, 对于一篇文章而言是好的处理方式
fx = feature_column.categorical_column_with_vocabulary_list(
'x', ['a', 'b', 'c', 'd'], dtype=tf.string, default_value=0)
fex = feature_column.embedding_column(fx, 4, 'mean')
t = feature_column.input_layer(x, [fex])
with tf.Session() as sess:
sess.run(tf.global_variables_initializer())
sess.run(tf.tables_initializer())
print(sess.run(t))
