PolicySearch.py 文件源码

def __init__(self,params,params_task,X,model,policy):

        self.rng = np.random.RandomState()

        self.model = model
        self.policy = policy

        self.params = params
        self.params_task = params_task


        self.x = T.matrix('x')
        cost  =  self.control(self.x)

        self.fwpass  = theano.function(inputs=[self.x], outputs = cost,allow_input_downcast=True)
        self.train_func = theano.function(inputs=[self.x],outputs=[cost], updates=self.adam(cost,lasagne.layers.get_all_params(self.policy,trainable=True),learning_rate=self.params['learning_rate']))

        self.policy_network = theano.function(inputs=[self.x],outputs=self.predict(self.x))