Python &引用;ValueError:未知层:&引用;使用Tensorflow调用copy.deepcopy(网络)时
我目前正在Tensorflow中设计一个NoiseNet,为此我需要定义一个自定义层。复制包含该自定义层的模型时,python会引发错误Python &引用;ValueError:未知层:&引用;使用Tensorflow调用copy.deepcopy(网络)时,python,python-3.x,tensorflow,valueerror,Python,Python 3.x,Tensorflow,Valueerror,我目前正在Tensorflow中设计一个NoiseNet,为此我需要定义一个自定义层。复制包含该自定义层的模型时,python会引发错误ValueError:Unknown layer:NoisyLayer。提供了该层的实现 目标是复制一个网络,创建它的第二个实例。为此,我使用命令net\u copy=copy.deepcopy(net\u original),只要不在要复制的模型中包含上面提到的自定义层,该命令就可以工作。 我看到,对于保存和加载,有一种方法可以指定自定义属性(例如自定义层),
ValueError:Unknown layer:NoisyLayer
。提供了该层的实现
目标是复制一个网络,创建它的第二个实例。为此,我使用命令net\u copy=copy.deepcopy(net\u original)
,只要不在要复制的模型中包含上面提到的自定义层,该命令就可以工作。
我看到,对于保存和加载,有一种方法可以指定自定义属性(例如自定义层),但我找不到类似的命令可以用于copy.deepcopy()
,其中copy是通过import copy
导入的
我正在Python3中使用Tensorflow 1.12.0
同样,在上面的链接下提供了自定义层。
使用自定义图层的网络如下所示:
class Network:
def __init__(self, actionspace_size, learning_rate, gradient_momentum, gradient_min):
frames_input = keras.layers.Input((84, 84, 4))
actions_input = keras.layers.Input((actionspace_size,))
conv1 = keras.layers.Conv2D(16, (8, 8), strides=(4, 4), activation="relu")(frames_input)
conv2 = keras.layers.Conv2D(32, (4, 4), strides=(2, 2), activation="relu")(conv1)
flattened = keras.layers.Flatten()(conv2)
# NoisyNet
hidden = NoisyLayer(activation=tf.nn.relu)(inputs=flattened, resample_noise_flag=True)
output = NoisyLayer(in_shape=(1,256), out_units=actionspace_size)(inputs=hidden, resample_noise_flag=True)
filtered_output = keras.layers.merge.Multiply()([output, actions_input])
self.model = keras.models.Model(inputs=[frames_input, actions_input], outputs=filtered_output)
self.model.compile(loss='mse', optimizer=keras.optimizers.RMSprop(lr=learning_rate, rho=gradient_momentum, epsilon=gradient_min))
# Import
import copy
from keras.utils import CustomObjectScope
# Copy
with CustomObjectScope({"MyCustomLayer":MyCustomLayer}):
model_copy = copy.deepcopy(model)
打电话的时候
q_net = Network(actionspace_size, learning_rate, gradient_momentum, gradient_min).
target_net = copy.deepcopy(q_net)
出现以下错误:
Traceback (most recent call last):
File "DQN_tf_NoisyNet.py", line 315, in <module>
main()
File "DQN_tf_NoisyNet.py", line 252, in main
target_net = copy.deepcopy(q_net)
File "/usr/lib/python3.5/copy.py", line 182, in deepcopy
y = _reconstruct(x, rv, 1, memo)
File "/usr/lib/python3.5/copy.py", line 299, in _reconstruct
y.__setstate__(state)
File "/usr/local/lib/python3.5/dist-packages/keras/engine/network.py", line 1266, in __setstate__
model = saving.unpickle_model(state)
File "/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py", line 435, in unpickle_model
return _deserialize_model(f)
File "/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py", line 225, in _deserialize_model
model = model_from_config(model_config, custom_objects=custom_objects)
File "/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py", line 458, in model_from_config
return deserialize(config, custom_objects=custom_objects)
File "/usr/local/lib/python3.5/dist-packages/keras/layers/__init__.py", line 55, in deserialize
printable_module_name='layer')
File "/usr/local/lib/python3.5/dist-packages/keras/utils/generic_utils.py", line 145, in deserialize_keras_object
list(custom_objects.items())))
File "/usr/local/lib/python3.5/dist-packages/keras/engine/network.py", line 1022, in from_config
process_layer(layer_data)
File "/usr/local/lib/python3.5/dist-packages/keras/engine/network.py", line 1008, in process_layer
custom_objects=custom_objects)
File "/usr/local/lib/python3.5/dist-packages/keras/layers/__init__.py", line 55, in deserialize
printable_module_name='layer')
File "/usr/local/lib/python3.5/dist-packages/keras/utils/generic_utils.py", line 138, in deserialize_keras_object
': ' + class_name)
ValueError: Unknown layer: NoisyLayer
回溯(最近一次呼叫最后一次):
文件“DQN_tf_noisenet.py”,第315行,在
main()
文件“DQN_tf_noisenet.py”,第252行,主目录
target\u net=copy.deepcopy(q\u net)
文件“/usr/lib/python3.5/copy.py”,第182行,在deepcopy中
y=_(x,rv,1,备忘录)
文件“/usr/lib/python3.5/copy.py”,第299行,在
y、 设置状态(状态)
文件“/usr/local/lib/python3.5/dist-packages/keras/engine/network.py”,第1266行,在__
模型=保存。取消勾选模型(状态)
文件“/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py”,第435行,在unpickle_模型中
返回反序列化模型(f)
文件“/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py”,第225行,在反序列化模型中
模型=来自配置的模型(模型配置,自定义对象=自定义对象)
文件“/usr/local/lib/python3.5/dist-packages/keras/engine/saving.py”,第458行,型号为
返回反序列化(配置,自定义对象=自定义对象)
文件“/usr/local/lib/python3.5/dist-packages/keras/layers/_-init__.py”,第55行,反序列化
可打印\u模块\u name='layer')
文件“/usr/local/lib/python3.5/dist packages/keras/utils/generic_utils.py”,第145行,反序列化_keras_对象
列表(自定义对象.项())
文件“/usr/local/lib/python3.5/dist packages/keras/engine/network.py”,第1022行,from_config
处理层(层数据)
文件“/usr/local/lib/python3.5/dist-packages/keras/engine/network.py”,第1008行,进程层中
自定义对象=自定义对象)
文件“/usr/local/lib/python3.5/dist-packages/keras/layers/_-init__.py”,第55行,反序列化
可打印\u模块\u name='layer')
文件“/usr/local/lib/python3.5/dist packages/keras/utils/generic_utils.py”,第138行,反序列化_keras_对象
“:”+类名)
ValueError:未知层:NoisyLayer
我知道网络本身不是问题(deepcopy方法也不是),因为只要我用标准密集层替换NoisyLayers(自定义)就可以很好地工作
有人知道如何复制包含自定义图层的Tensorflow模型吗?提前谢谢 找到了一个解决方案:
问题是Tensorflow/Keras不知道如何解释自定义层。因此,为了提供如何解释层的信息,可以使用Keras的CustomObjectScope
并在该范围内复制模型,如下所示:
class Network:
def __init__(self, actionspace_size, learning_rate, gradient_momentum, gradient_min):
frames_input = keras.layers.Input((84, 84, 4))
actions_input = keras.layers.Input((actionspace_size,))
conv1 = keras.layers.Conv2D(16, (8, 8), strides=(4, 4), activation="relu")(frames_input)
conv2 = keras.layers.Conv2D(32, (4, 4), strides=(2, 2), activation="relu")(conv1)
flattened = keras.layers.Flatten()(conv2)
# NoisyNet
hidden = NoisyLayer(activation=tf.nn.relu)(inputs=flattened, resample_noise_flag=True)
output = NoisyLayer(in_shape=(1,256), out_units=actionspace_size)(inputs=hidden, resample_noise_flag=True)
filtered_output = keras.layers.merge.Multiply()([output, actions_input])
self.model = keras.models.Model(inputs=[frames_input, actions_input], outputs=filtered_output)
self.model.compile(loss='mse', optimizer=keras.optimizers.RMSprop(lr=learning_rate, rho=gradient_momentum, epsilon=gradient_min))
# Import
import copy
from keras.utils import CustomObjectScope
# Copy
with CustomObjectScope({"MyCustomLayer":MyCustomLayer}):
model_copy = copy.deepcopy(model)
这负责复制部分。但是,只要没有将自定义输入指定为自定义层的构造函数的参数(\uuu init(…)
),就可以立即执行此操作
我猜是这样的,因为在幕后,copy()函数似乎临时保存原始模型,然后使用一些pickle
-功能再次加载原始模型,因此必须声明更多构造函数参数的值,如下所示:
class Network:
def __init__(self, actionspace_size, learning_rate, gradient_momentum, gradient_min):
frames_input = keras.layers.Input((84, 84, 4))
actions_input = keras.layers.Input((actionspace_size,))
conv1 = keras.layers.Conv2D(16, (8, 8), strides=(4, 4), activation="relu")(frames_input)
conv2 = keras.layers.Conv2D(32, (4, 4), strides=(2, 2), activation="relu")(conv1)
flattened = keras.layers.Flatten()(conv2)
# NoisyNet
hidden = NoisyLayer(activation=tf.nn.relu)(inputs=flattened, resample_noise_flag=True)
output = NoisyLayer(in_shape=(1,256), out_units=actionspace_size)(inputs=hidden, resample_noise_flag=True)
filtered_output = keras.layers.merge.Multiply()([output, actions_input])
self.model = keras.models.Model(inputs=[frames_input, actions_input], outputs=filtered_output)
self.model.compile(loss='mse', optimizer=keras.optimizers.RMSprop(lr=learning_rate, rho=gradient_momentum, epsilon=gradient_min))
# Import
import copy
from keras.utils import CustomObjectScope
# Copy
with CustomObjectScope({"MyCustomLayer":MyCustomLayer}):
model_copy = copy.deepcopy(model)
如果自定义类的开头如下所示,其中output\u dim
是上述自定义参数之一:
class MyCustomLayer(keras.layers.Layer):
def __init__(self, output_dim, **kwargs):
self.output_dim = output_dim
super(MyCustomLayer, self).__init__(**kwargs)
然后必须向类MyCustomLayer
添加一个函数,该函数还负责保存和加载自定义构造函数参数(复制时):
这两个步骤解决了我的问题。找到了解决方案:
问题是Tensorflow/Keras不知道如何解释自定义层。因此,为了提供如何解释层的信息,可以使用Keras的CustomObjectScope
并在该范围内复制模型,如下所示:
class Network:
def __init__(self, actionspace_size, learning_rate, gradient_momentum, gradient_min):
frames_input = keras.layers.Input((84, 84, 4))
actions_input = keras.layers.Input((actionspace_size,))
conv1 = keras.layers.Conv2D(16, (8, 8), strides=(4, 4), activation="relu")(frames_input)
conv2 = keras.layers.Conv2D(32, (4, 4), strides=(2, 2), activation="relu")(conv1)
flattened = keras.layers.Flatten()(conv2)
# NoisyNet
hidden = NoisyLayer(activation=tf.nn.relu)(inputs=flattened, resample_noise_flag=True)
output = NoisyLayer(in_shape=(1,256), out_units=actionspace_size)(inputs=hidden, resample_noise_flag=True)
filtered_output = keras.layers.merge.Multiply()([output, actions_input])
self.model = keras.models.Model(inputs=[frames_input, actions_input], outputs=filtered_output)
self.model.compile(loss='mse', optimizer=keras.optimizers.RMSprop(lr=learning_rate, rho=gradient_momentum, epsilon=gradient_min))
# Import
import copy
from keras.utils import CustomObjectScope
# Copy
with CustomObjectScope({"MyCustomLayer":MyCustomLayer}):
model_copy = copy.deepcopy(model)
这负责复制部分。但是,只要没有将自定义输入指定为自定义层的构造函数的参数(\uuu init(…)
),就可以立即执行此操作
我猜是这样的,因为在幕后,copy()函数似乎临时保存原始模型,然后使用一些pickle
-功能再次加载原始模型,因此必须声明更多构造函数参数的值,如下所示:
class Network:
def __init__(self, actionspace_size, learning_rate, gradient_momentum, gradient_min):
frames_input = keras.layers.Input((84, 84, 4))
actions_input = keras.layers.Input((actionspace_size,))
conv1 = keras.layers.Conv2D(16, (8, 8), strides=(4, 4), activation="relu")(frames_input)
conv2 = keras.layers.Conv2D(32, (4, 4), strides=(2, 2), activation="relu")(conv1)
flattened = keras.layers.Flatten()(conv2)
# NoisyNet
hidden = NoisyLayer(activation=tf.nn.relu)(inputs=flattened, resample_noise_flag=True)
output = NoisyLayer(in_shape=(1,256), out_units=actionspace_size)(inputs=hidden, resample_noise_flag=True)
filtered_output = keras.layers.merge.Multiply()([output, actions_input])
self.model = keras.models.Model(inputs=[frames_input, actions_input], outputs=filtered_output)
self.model.compile(loss='mse', optimizer=keras.optimizers.RMSprop(lr=learning_rate, rho=gradient_momentum, epsilon=gradient_min))
# Import
import copy
from keras.utils import CustomObjectScope
# Copy
with CustomObjectScope({"MyCustomLayer":MyCustomLayer}):
model_copy = copy.deepcopy(model)
如果自定义类的开头如下所示,其中output\u dim
是上述自定义参数之一:
class MyCustomLayer(keras.layers.Layer):
def __init__(self, output_dim, **kwargs):
self.output_dim = output_dim
super(MyCustomLayer, self).__init__(**kwargs)
然后必须向类MyCustomLayer
添加一个函数,该函数还负责保存和加载自定义构造函数参数(复制时):
这两个步骤解决了我的问题。是否有其他方法可以避免复制整个网络,只复制等效网络架构之间的单个权重和偏差?是否有其他方法可以避免复制整个网络,而是在等效网络架构之间复制单个权重和偏差?