Created by: maosengshulei

单机训练时：

vars = list(self.learn_program.list_vars())

    target_vars = list(
        filter(lambda x: 'GRAD' not in x.name and 'target' in x.name, vars))
    policy_vars_name = [
        x.name.replace('target', 'policy') for x in target_vars
    ]
    policy_vars = list(filter(lambda x: x.name in policy_vars_name, vars))

    policy_vars.sort(key=lambda x: x.name)
    target_vars.sort(key=lambda x: x.name)
    
    with fluid.program_guard(self._sync_program):
        sync_ops = []
        decay = 1.0 - TAU
        for i, var in enumerate(policy_vars):
            #sync_op = fluid.layers.assign(decay * target_vars[i] + (1 - decay) * policy_vars[i], target_vars[i])
            sync_op = fluid.layers.assign(policy_vars[i], target_vars[i])
            sync_ops.append(sync_op)`

训练正常。 当我想把这种hard target updates 的方式换成soft target updates 即：

    with fluid.program_guard(self._sync_program):
        sync_ops = []
        decay = 1.0 - TAU
        for i, var in enumerate(policy_vars):
            sync_op = fluid.layers.assign(decay * target_vars[i] + (1 - decay) * policy_vars[i], target_vars[i])
            #sync_op = fluid.layers.assign(policy_vars[i], target_vars[i])
            sync_ops.append(sync_op)`          

训练时会报错： Error: The Input Variable(X) of assign Op used to determine kernel data type is empty or not LoDTensor or SelectedRows. [Hint: Expected data_type != dafault_data_type, but received data_type:-1 == dafault_data_type:-1.] at (/home/teamcity/work/ef54dc8a5b211854/paddle/fluid/framework/operator.cc:1303) [operator < assign > error]

请教一下错误的原因，和实现soft target update的方法