def __init__(self):
self.fc1 = layers.fc(size=512, act='relu', name='fc1')
self.fc2 = layers.fc(size=512, act='relu', name='fc2')
self.fc3 = layers.fc(size=512, act='relu', name='fc3')
self.fc4 = layers.fc(size=512, act='relu', name='fc4')
self.fc5 = layers.fc(size=512, act='relu', name='fc5')
self.fc6 = layers.fc(size=500, act='relu', name='fc6')
self.month_embedding = layers.embedding(size=[12, 64],
name='emb_month')
self.hour_embedding = layers.embedding(size=[24, 64], name='emb_hour')
def __init__(self):
self.fc1 = layers.fc(size=1024, act='relu', name='fc1')
self.fc2 = layers.fc(size=900, act='relu', name='fc2')
self.fc3 = layers.fc(size=800, act='relu', name='fc3')
self.fc4 = layers.fc(size=700, act='relu', name='fc4')
self.fc5 = layers.fc(size=512, act='relu', name='fc5')
self.fc6 = layers.fc(size=label_dim, act='relu', name='fc6')
self.month_embedding = layers.embedding(size=[13, 256],
name='emb_month')
self.hour_embedding = layers.embedding(size=[24, 256], name='emb_hour')
self.line_status_embedding = layers.embedding(size=[400, 256],
name='emb_line_status')
def __init__(self, act_dim):
hid1_size = 256
hid2_size = 256
hid3_size = 256
# 3层全连接网络
self.emb_1 = layers.embedding(size=[128, 64])
self.fc1 = layers.fc(size=hid1_size, act='relu')
self.fc2 = layers.fc(size=hid2_size, act='relu')
self.fc4 = layers.fc(size=act_dim, act=None)
def __init__(self):
self.fc1 = layers.fc(100)
self.fc2 = layers.fc(100)
self.fc3 = layers.fc(100, bias_attr=False)
self.fc4 = layers.fc(100, param_attr=False)
self.fc5 = layers.fc(100, name="fc", bias_attr=False)
self.fc6 = layers.fc(100, param_attr=fluid.initializer.Xavier())
self.embedding = layers.embedding((100, 128))
self.embedding_custom = layers.embedding((100, 128),
name="embedding_custom")
## although here self.conv2d shares param with self.embedding,
## it might be invalid because the param sizes do not match
self.conv2d = layers.conv2d(
num_filters=64,
filter_size=3,
param_attr=self.embedding.attr_holder.param_attr,
name="my_conv2d")
self.batch_norm = layers.batch_norm()
def default_embedding(size, name):
gradient_clip = default_param_clip()
reg = fluid.regularizer.L2Decay(1e-5) # IMPORTANT, to prevent overfitting.
embed = layers.embedding(name=name,
size=size,
param_attr=ParamAttr(
initializer=fluid.initializer.Xavier(),
gradient_clip=gradient_clip,
regularizer=reg),
is_sparse=False)
return embed
def default_embedding(size, name, embed_clip, regularizer=None):
gradient_clip = default_param_clip() if embed_clip else None
embed = layers.embedding(name=name,
size=size,
param_attr=ParamAttr(initializer=fluid.initializer.Xavier(),
gradient_clip=gradient_clip,
regularizer=regularizer),
is_sparse=False, # turn on lazy_mode when using Adam
is_distributed=False, # TODO https://github.com/PaddlePaddle/Paddle/issues/15133
)
return embed
def __init__(self):
self.fc1 = layers.fc(64, bias_attr=False)
self.fc2 = layers.fc(64, bias_attr=False)
self.fc3 = layers.fc(64, name="fc")
self.fc4 = layers.fc(64, name="fc")
self.embedding = layers.embedding(
(100, 64), param_attr=self.fc1.attr_holder.param_attr)
self.created_param = layers.create_parameter(
shape=[100],
dtype='float32',
default_initializer=fluid.initializer.Uniform(low=-1.0, high=1.0))
self.batch_norm = layers.batch_norm()