Question

不知道该在哪里发布，所以...

我怀疑我的Tensorforce代理未按预期运行...每次尝试提高屏幕快照（状态）分辨率时，它都会在RTX 2060 GPU上显示ResourceExhaustedError ...我想改善我的代码的效率，但是真的不知道该怎么做。

代码：

def agent_build():
    #try:
    #    agent = DQN.restore_model(join(getcwd(), "models"))
    #except:
    if True:
        network = [
            #dict(type='conv1d', size=256, window=1, stride=1),  
            dict(type='flatten'),
            dict(type='dueling', size=256, activation='swish'),
            dict(type='dense', size=256, activation='swish')
            ]

        print("Creating agent.")
        agent = DQN(
            states=dict(type='float', #Int may be more appropriate...
                        shape=(screen_size()[1], screen_size()[0])),
            actions=dict(type='int', num_actions=2000),
            network=network,
            update_mode = dict(units='episodes', batch_size=2, frequency=2),
            batching_capacity = 256
        )
    return agent

def smart_rewarder_build():
    network = [
        dict(type='flatten'),
        dict(type='dense', size=64, activation='swish'),
        dict(type='dense', size=64, activation='swish')
            ]
    agent = DQN(
        states = dict(type='float', #Int may be more appropriate...
                        shape=(screen_size()[1], screen_size()[0])),
        actions = dict(type = 'int', num_actions=11), # -5...10
        network=network
        )
return agent

错误消息：

    2019-03-03 05:51:39.337780: W tensorflow/core/framework
/op_kernel.cc:1273] OP_REQUIRES failed at random_op.cc:202 : Resource 
exhausted: OOM when allocating tensor with shape[32000,162,288] and type 
float on /job:localhost/replica:0/task:0/device:GPU:0 by allocator 
GPU_0_bfc

我的Tensorforce代理配置正确吗？

0 个答案: