I have trained a Tensorflow model and saved it on another machine and want to load it locally. When i try to load it i get an error saying Agent.init() got an unexpected keyword argument 'name'. My Agent class is the neural net i want to load but no keyword called name is passed to it. My Agent class code is:
class Agent(Model):
"""
Defines a class for the actors used in reinforcement leraning where the states are represented as a 2-D image
params:
number_of_outputs: the number of outputs the neural net should return
number_of_hidden_units: the number of hidden units in the neural net
"""
def __init__(self,number_of_outputs: int,number_of_hidden_units: int):
super(Agent,self).__init__()
self.number_of_outputs = number_of_outputs
self.number_of_hidden_units = number_of_hidden_units
self.first_block = Sequential(
[
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu',data_format = 'channels_last', kernel_initializer='he_normal'),
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu',data_format = 'channels_last', kernel_initializer='he_normal'),
MaxPooling2D(pool_size=3, padding='same')
]
)
self.second_block = Sequential(
[
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu', data_format = 'channels_last', kernel_initializer='he_normal'),
MaxPooling2D(pool_size=3, padding='same')
]
)
self.prediction_block = Sequential(
[
Flatten(),
Dense(128,activation = 'linear'),
Dense(number_of_outputs, activation = 'linear')
]
)
self.relu = ReLU()
self.dropout = Dropout(0.25)
self.normalize = BatchNormalization()
def call(self,data):
x = self.first_block(data)
x = self.normalize(x)
x = self.second_block(x)
x = self.normalize(x)
x = self.prediction_block(x)
return x
def get_config(self):
base_config = super().get_config()
config = {
"number_of_outputs": self.number_of_outputs,
"number_of_hidden_units" :self.number_of_hidden_units
}
return {**base_config, **config}
The code used to save the model is:
def save_full_model(self, episode):
self.model.save(f'dqn_model_{episode}.h5')
And the code used to load the saved model is:
def load_full_model(self, path_to_model):
self.model = load_model(path_to_model, custom_objects = {'Agent':Agent} )
Note that the exact same Agent class was used during training.
I have trained a Tensorflow model and saved it on another machine and want to load it locally. When i try to load it i get an error saying Agent.init() got an unexpected keyword argument 'name'. My Agent class is the neural net i want to load but no keyword called name is passed to it. My Agent class code is:
class Agent(Model):
"""
Defines a class for the actors used in reinforcement leraning where the states are represented as a 2-D image
params:
number_of_outputs: the number of outputs the neural net should return
number_of_hidden_units: the number of hidden units in the neural net
"""
def __init__(self,number_of_outputs: int,number_of_hidden_units: int):
super(Agent,self).__init__()
self.number_of_outputs = number_of_outputs
self.number_of_hidden_units = number_of_hidden_units
self.first_block = Sequential(
[
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu',data_format = 'channels_last', kernel_initializer='he_normal'),
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu',data_format = 'channels_last', kernel_initializer='he_normal'),
MaxPooling2D(pool_size=3, padding='same')
]
)
self.second_block = Sequential(
[
Conv2D(number_of_hidden_units, kernel_size=2, padding='same', strides=1, activation = 'relu', data_format = 'channels_last', kernel_initializer='he_normal'),
MaxPooling2D(pool_size=3, padding='same')
]
)
self.prediction_block = Sequential(
[
Flatten(),
Dense(128,activation = 'linear'),
Dense(number_of_outputs, activation = 'linear')
]
)
self.relu = ReLU()
self.dropout = Dropout(0.25)
self.normalize = BatchNormalization()
def call(self,data):
x = self.first_block(data)
x = self.normalize(x)
x = self.second_block(x)
x = self.normalize(x)
x = self.prediction_block(x)
return x
def get_config(self):
base_config = super().get_config()
config = {
"number_of_outputs": self.number_of_outputs,
"number_of_hidden_units" :self.number_of_hidden_units
}
return {**base_config, **config}
The code used to save the model is:
def save_full_model(self, episode):
self.model.save(f'dqn_model_{episode}.h5')
And the code used to load the saved model is:
def load_full_model(self, path_to_model):
self.model = load_model(path_to_model, custom_objects = {'Agent':Agent} )
Note that the exact same Agent class was used during training.
Share Improve this question edited Mar 14 at 16:12 desertnaut 60.5k32 gold badges155 silver badges182 bronze badges asked Mar 12 at 23:29 borjanobborjanob 151 silver badge4 bronze badges1 Answer
Reset to default 2This is likely happening because you overwrite __init__ in a way that does not accept the base Model class arguments like name, but use get_config from the super class. So from the config, the class expects to have the usual Model arguments like name. This should be fixed by adding the necessary arguments to __init__, easiest with **kwargs:
def __init__(self,number_of_outputs: int,number_of_hidden_units: int, **kwargs):
super(Agent,self).__init__(**kwargs)
发布者：admin，转转请注明出处：http://www.yc00.com/questions/1744726110a4590188.html
评论列表（0条）