def __init__(self,
model_config,
inputs,
labels,
scope='LR',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
self.mode = mode
with tf.variable_scope(scope, default_name='embeddings'):
linear_dense_value_list, linear_sparse_embedding_list = input_from_feature_columns(
self.inputs,
self.config.linear_feature_columns,
target='linear')
linear_logits = Linear()(linear_dense_value_list,
linear_sparse_embedding_list)
self.logits = linear_logits
if self.mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
if len(labels.shape) == 1:
labels = tf.expand_dims(labels, axis=-1)
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels) +
tf.losses.get_regularization_loss())
def __init__(self,
model_config,
inputs,
labels,
scope='FFM',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
with tf.variable_scope(scope, default_name='embeddings'):
_, ffm_sparse_embedding_dict = input_from_feature_columns(
self.inputs,
self.config.linear_feature_columns,
target='dnn',
field=True)
linear_dense_value_list, linear_sparse_embedding_list = input_from_feature_columns(
self.inputs,
self.config.linear_feature_columns,
target='linear',
field=False)
linear_logits = Linear()(linear_dense_value_list,
linear_sparse_embedding_list)
self.logits = linear_logits
sparse_feature_columns = [
fc for fc in self.config.linear_feature_columns
if isinstance(fc, SparseFeature)
]
if len(sparse_feature_columns) > 1:
ffm_logits = self.ffm(sparse_feature_columns,
ffm_sparse_embedding_dict)
self.logits += ffm_logits
if mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
if len(labels.shape) == 1:
labels = tf.expand_dims(labels, axis=-1)
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels))
regularity_loss = tf.losses.get_regularization_loss()
# tf.losses.get_regularization_losses 获取的应该是list， tf.losses.get_regularization_loss是加过的
self.loss = self.loss + regularity_loss
def __init__(self,
model_config,
inputs,
labels,
scope='PNN',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
self.mode = mode
self.use_bn = model_config.use_bn
self.method = model_config.method
with tf.variable_scope(scope, default_name='embeddings'):
dnn_dense_value_list, dnn_sparse_embedding_list = input_from_feature_columns(
self.inputs, self.config.dnn_feature_columns, target='dnn')
for i in range(len(dnn_dense_value_list)):
if len(dnn_dense_value_list[i].shape) == 1:
dnn_dense_value_list[i] = tf.expand_dims(
dnn_dense_value_list[i], axis=-1)
linear_signal = tf.concat(dnn_dense_value_list +
dnn_sparse_embedding_list,
axis=-1)
dnn_input = linear_signal
if len(dnn_sparse_embedding_list) > 0:
inner_product_signal = self.inner_product(
dnn_sparse_embedding_list)
dnn_input = tf.concat([dnn_input, linear_signal], axis=-1)
dnn_out = DNN(units=self.config.units,
activation=self.config.activation,
dropout_rate=self.config.dropout_rate,
use_bn=self.config.use_bn,
training=mode == tf.estimator.ModeKeys.TRAIN,
toonedim=False)([], [dnn_input])
# hidden = dnn_input
# for u in self.config.units:
# hidden =tf.layers.dense(hidden, u, kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))
# if self.config.use_bn:
# hidden = tf.layers.batch_normalization(hidden, training=mode==tf.estimator.ModeKeys.TRAIN)
# hidden = tf.layers.dropout(hidden, rate=0.5, training=mode==tf.estimator.ModeKeys.TRAIN)
self.logits = tf.layers.dense(
dnn_out,
1,
kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))
self.logits = tf.reduce_sum(self.logits, axis=-1)
if self.mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels) +
tf.losses.get_regularization_loss())
def __init__(self,
model_config,
inputs,
labels,
scope='WDL',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
self.mode = mode
self.use_bn = model_config.use_bn
with tf.variable_scope(scope, default_name='embeddings'):
linear_dense_value_list, linear_sparse_embedding_list = input_from_feature_columns(
self.inputs,
self.config.linear_feature_columns,
target='linear')
dnn_dense_value_list, dnn_sparse_embedding_list = input_from_feature_columns(
self.inputs, self.config.dnn_feature_columns, target='dnn')
linear_logits = Linear()(linear_dense_value_list,
linear_sparse_embedding_list)
dnn_logits = DNN(units=self.config.units,
activation=self.config.activation,
dropout_rate=self.config.dropout_rate,
use_bn=True,
training=tf.estimator.ModeKeys.TRAIN == mode,
toonedim=True)(dnn_dense_value_list,
dnn_sparse_embedding_list)
self.logits = linear_logits + dnn_logits
if self.mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
if len(labels.shape) == 1:
labels = tf.expand_dims(labels, axis=-1)
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels) +
tf.losses.get_regularization_loss())
def __init__(self,
model_config,
inputs,
labels,
scope='DCN',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
self.mode = mode
self.use_bn = model_config.use_bn
self.num_crosses = model_config.num_crosses
with tf.variable_scope(scope, default_name='embeddings'):
dnn_dense_value_list, dnn_sparse_embedding_list = input_from_feature_columns(
self.inputs, self.config.dnn_feature_columns, target='dnn')
if len(dnn_sparse_embedding_list) > 0:
cross_logits = self.cross(dnn_sparse_embedding_list)
deep_logits = DNN(units=self.config.units,
activation=self.config.activation,
dropout_rate=self.config.dropout_rate,
use_bn=self.config.use_bn,
training=mode == tf.estimator.ModeKeys.TRAIN,
toonedim=False)(dnn_dense_value_list,
dnn_sparse_embedding_list)
logits = tf.concat([cross_logits, deep_logits], axis=-1)
else:
deep_logits = DNN(units=self.config.units,
activation=self.config.activation,
dropout_rate=self.config.dropout_rate,
use_bn=self.config.use_bn,
training=mode == tf.estimator.ModeKeys.TRAIN,
toonedim=False)(dnn_dense_value_list,
dnn_sparse_embedding_list)
logits = deep_logits
self.logits = tf.layers.dense(
logits,
1,
kernel_regularizer=tf.contrib.layers.l2_regularizer(0.001))
self.logits = tf.reduce_sum(self.logits, axis=-1)
if self.mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels) +
tf.losses.get_regularization_loss())
def __init__(self,
model_config,
inputs,
labels,
scope='MLR',
mode=tf.estimator.ModeKeys.TRAIN):
self.config = copy.deepcopy(model_config)
self.inputs = inputs
self.labels = labels
self.mode = mode
self.num_lr = model_config.num_lr
with tf.variable_scope(scope, default_name='embeddings'):
linear_dense_value_list, linear_sparse_embedding_list = input_from_feature_columns(
self.inputs,
self.config.linear_feature_columns,
target='linear')
lr_logits_list = []
for i in range(self.num_lr):
with tf.variable_scope('lr%d' % i, default_name='linear'):
lr_logits = Linear()(linear_dense_value_list,
linear_sparse_embedding_list)
lr_logits_list.append(lr_logits)
logits = tf.concat(lr_logits_list, axis=-1)
self.logits = tf.layers.dense(logits, 1, name='logits_weight')
self.logits = tf.reduce_mean(self.logits, axis=-1)
if self.mode == tf.estimator.ModeKeys.PREDICT:
self.loss = None
else:
self.loss = tf.reduce_mean(
tf.nn.sigmoid_cross_entropy_with_logits(logits=self.logits,
labels=labels) +
tf.losses.get_regularization_loss())
def DeepFM(linear_feature_columns,
dnn_feature_columns,
embedding_size=8,
use_fm=True,
dnn_hidden_units=(128, 128),
l2_reg_linear=0.00001,
l2_reg_embedding=0.00001,
l2_reg_dnn=0,
init_std=0.0001,
seed=1024,
dnn_dropout=0,
dnn_activation='relu',
dnn_use_bn=False,
task='binary',
att=False,
seq_len=None,
cate_feats=[],
cate2nunique={}):
"""Instantiates the DeepFM Network architecture.
:param linear_feature_columns: An iterable containing all the features used by linear part of the model.
:param dnn_feature_columns: An iterable containing all the features used by deep part of the model.
:param embedding_size: positive integer,sparse feature embedding_size
:param use_fm: bool,use FM part or not
:param dnn_hidden_units: list,list of positive integer or empty list, the layer number and units in each layer of DNN
:param l2_reg_linear: float. L2 regularizer strength applied to linear part
:param l2_reg_embedding: float. L2 regularizer strength applied to embedding vector
:param l2_reg_dnn: float. L2 regularizer strength applied to DNN
:param init_std: float,to use as the initialize std of embedding vector
:param seed: integer ,to use as random seed.
:param dnn_dropout: float in [0,1), the probability we will drop out a given DNN coordinate.
:param dnn_activation: Activation function to use in DNN
:param dnn_use_bn: bool. Whether use BatchNormalization before activation or not in DNN
:param task: str, ``"binary"`` for binary logloss or ``"regression"`` for regression loss
:return: A Keras model instance.
"""
features = build_input_features(linear_feature_columns +
dnn_feature_columns)
inputs_list = list(features.values())
sparse_embedding_list, dense_value_list, embedding_dict = input_from_feature_columns(
features, dnn_feature_columns, embedding_size, l2_reg_embedding,
init_std, seed)
linear_logit = get_linear_logit(features,
linear_feature_columns,
l2_reg=l2_reg_linear,
init_std=init_std,
seed=seed,
prefix='linear')
fm_input = concat_fun(sparse_embedding_list, axis=1)
fm_logit = FM()(fm_input)
dnn_input = combined_dnn_input(sparse_embedding_list, dense_value_list)
input_lstm = Input(shape=(seq_len, 1 + len(cate_feats)), name='lstm_input')
input_lstm_gap = Lambda(lambda x: x[:, :, 0:1])(input_lstm)
concate_list = [input_lstm_gap]
for i, cate in enumerate(cate_feats):
input_cate = Lambda(lambda x: x[:, :, i + 1])(input_lstm)
emb = embedding_dict.get(cate)
if emb is None:
emb = Embedding(output_dim=8, input_dim=cate2nunique[cate])
concate_list.append(emb(input_cate))
input_lstm_concat = Concatenate(axis=-1)(concate_list)
if att:
lstm_out = LSTM(units=128, return_sequences=True)(input_lstm_concat)
attention_mul = attention_3d_block(lstm_out, seq_len)
lstm_out = Lambda(lambda x: K.sum(x, axis=1))(attention_mul)
else:
lstm_out = LSTM(units=128, return_sequences=False)(input_lstm_concat)
dnn_input = concat_fun([dnn_input, lstm_out])
dnn_out = DNN(dnn_hidden_units, dnn_activation, l2_reg_dnn, dnn_dropout,
dnn_use_bn, seed)(dnn_input)
dnn_logit = tf.keras.layers.Dense(1, use_bias=False,
activation=None)(dnn_out)
if len(dnn_hidden_units) == 0 and use_fm == False: # only linear
final_logit = linear_logit
elif len(dnn_hidden_units) == 0 and use_fm == True: # linear + FM
final_logit = tf.keras.layers.add([linear_logit, fm_logit])
elif len(dnn_hidden_units) > 0 and use_fm == False: # linear + Deep
final_logit = tf.keras.layers.add([linear_logit, dnn_logit])
elif len(dnn_hidden_units) > 0 and use_fm == True: # linear + FM + Deep
final_logit = tf.keras.layers.add([linear_logit, fm_logit, dnn_logit])
else:
raise NotImplementedError
output = PredictionLayer(task)(final_logit)
model = tf.keras.models.Model(inputs=inputs_list + [input_lstm],
outputs=output)
return model
