Created by: Aurelius84

append_backward API输入类型检查（仅涉及python端）

1. loss参数必须为Variable

fluid.backward.append_backward(loss=avg_loss.name)  # 若传入string类型

• 优化前只有一个assert语句相关报错

  File "backward.py", line 1245, in append_backward
    assert isinstance(loss, framework.Variable)
AssertionError

• 优化后，会提示当前转入的类型

TypeError: The type of 'loss' in fluid.backward.append_backward must be <class 'paddle.fluid.framework.Variable'>, but received <class 'str'>.

2. parameter_list参数为list[Variable|str]

param_names = [
    param.name
    for param in fluid.default_main_program().block(0).all_parameters()
]
param_names[0] = np.random.random([10])
fluid.backward.append_backward(loss=avg_loss, parameter_list=self.param_names)

• 优化前，若parameter_list的元素不是Varibale或str，报如下错：

TypeError: The type of parameter_list's member must be paddle.fluid.Variable or str, but received <class 'numpy.ndarray'>.

• 优化后，增加了具体元素的位置信息

TypeError: The type of 'parameter_list[0]' in fluid.backward.append_backward must be (<class 'paddle.fluid.framework.Variable'>, (<class 'str'>,)), but received <class 'numpy.ndarray'>.

3. callbacks参数必须为list

def callback(block, context):
    return

fluid.backward.append_backward(loss=avg_loss, callbacks=callback)

• 优化前，仅会输出assert信息

_append_backward_ops_
    assert (isinstance(callbacks, list))
AssertionError

• 优化后，按照规范输出报错信息

TypeError: The type of 'callbacks' in fluid.backward.append_backward must be <class 'list'>, but received <class 'function'>.

• 移除了对fluid.gradients的no_grad_set的强制set类型检查

之前存在no_grad_set传入list、tuple的情况，已在内部进行兼容处理，并已经给出了完备的报错信息提示，故此处移除对fluid.gradients的no_grad_set的强制set类型检查