def create_model(self, model_info):
"""Create Deep-Q network."""
model_config = model_info['model_config']
zeus_model_name = model_config['zeus_model_name']
model_cls = ClassFactory.get_cls(ClassType.SEARCH_SPACE,
zeus_model_name)
zeus_model = model_cls(state_dim=self.state_dim,
action_dim=self.action_dim)
# zeus_model = DqnMlpNet(state_dim=self.state_dim, action_dim=self.action_dim)
LossConfig.type = model_config['loss']
OptimConfig.type = model_config['optim']
OptimConfig.params.update({'lr': model_config['lr']})
loss_input = dict()
loss_input['inputs'] = [{
"name": "input_state",
"type": "float32",
"shape": self.state_dim
}]
loss_input['labels'] = [{
"name": "target_value",
"type": "float32",
"shape": self.action_dim
}]
model = Trainer(model=zeus_model,
backend='tensorflow',
device='GPU',
loss_input=loss_input,
lazy_build=False)
return model
def create_model(self, model_info):
"""Create Deep-Q network."""
zeus_model = DqnMlpNet(state_dim=self.state_dim,
action_dim=self.action_dim)
LossConfig.type = 'mse_loss'
OptimConfig.type = 'Adam'
OptimConfig.params.update({'lr': self.learning_rate})
loss_input = dict()
loss_input['inputs'] = [{
"name": "input_state",
"type": "float32",
"shape": self.state_dim
}]
loss_input['labels'] = [{
"name": "target_value",
"type": "float32",
"shape": self.action_dim
}]
model = Trainer(model=zeus_model,
backend='tensorflow',
device='GPU',
loss_input=loss_input,
lazy_build=False)
return model
def create_model(self, model_info):
zeus_model = PpoMlpNet(state_dim=self.state_dim, action_dim=self.action_dim)
LossConfig.type = 'ppo_loss'
OptimConfig.type = 'Adam'
OptimConfig.params.update({'lr': LR})
loss_input = dict()
loss_input['inputs'] = [{"name": "input_state", "type": "float32", "shape": self.state_dim}]
loss_input['labels'] = [{"name": "old_v", "type": "float32", "shape": 1}]
loss_input['labels'].append({"name": "target_v", "type": "float32", "shape": 1})
loss_input['labels'].append({"name": "old_p", "type": "float32", "shape": self.action_dim})
loss_input['labels'].append({"name": "target_p", "type": "int32", "shape": 1})
loss_input['labels'].append({"name": "adv", "type": "float32", "shape": 1})
model = Trainer(model=zeus_model, lazy_build=False, loss_input=loss_input)
return model
def create_model(self, model_info):
"""Create zeus network."""
zeus_model = ImpalaMlpNet(state_dim=self.state_dim, action_dim=self.action_dim)
LossConfig.type = 'impala_loss'
OptimConfig.type = 'Adam'
OptimConfig.params.update({'lr': LR})
loss_input = dict()
loss_input['inputs'] = [{"name": "input_state", "type": "float32", "shape": self.state_dim}]
loss_input['labels'] = [{"name": "target_p", "type": "float32", "shape": self.action_dim}]
loss_input['labels'].append({"name": "target_v", "type": "float32", "shape": 1})
loss_input['labels'].append({"name": "adv", "type": "float32", "shape": 1})
model = Trainer(model=zeus_model, backend='tensorflow', device='GPU',
loss_input=loss_input, lazy_build=False)
return model