1 系统环境
硬件环境(Ascend/GPU/CPU): CPU
MindSpore版本: mindspore=1.10
执行模式（PyNative/ Graph）:不限
Python版本: Python=3.8
操作系统平台: linux
2 报错信息
2.1 问题描述
将model(x)放在dataset.create_dict_iterator()遍历的外面，程序正确运行，代码如下：
def test(dataset,model,mask_ratio=0.75):
batch_loss = 0
model.set_train(False)
i = 0
x = mindspore.Tensor(np.randon.rand(1, 40, 300).astype(np. float32), dtype=nindspore.float32)
loss, pred, mask = model(x, mask_ratio)
for data in dataset.create_dict_iterator():
# input = data['data']
# x = mindspore.Tensor(np.random.rand(1, 40, 300).astype(np.fLoat32), dtype=mindspore.float32)
# loss, pred, mask = model(x, mask_ratio)
batch_loss += loss
i += 1
return batch_loss/(i+1)
但是只要将model(x)放入dataset.create_dict_iterator()的遍历后，就会报错，即使x并不是dataset中的数据。
def test(dataset,model,mask_ratio=0.75):
batch_loss = 0
model.set_train(False)
i = 0
x = mindspore.Tensor(np.randon.rand(1, 40, 300).astype(np. float32), dtype=nindspore.float32)
# loss, pred, mask = model(x, mask_ratio)
for data in dataset.create_dict_iterator():
# input = data['data']
x = mindspore.Tensor(np.random.rand(1, 40, 300).astype(np.fLoat32), dtype=mindspore.float32)
loss, pred, mask = model(x, mask_ratio)
batch_loss += loss
i += 1
return batch_loss/(i+1)
2.2 报错信息
RuntimeError: Illegal AnfNode for evaluating.node: @BatchNorm.20:фdout(type:Parameter), fg: BatchNorm.20 conf: Node: 0x18b20bbdb00/@BatchNorm.20:фdout-uid(2115), Context: 0x18b20b4fc70/{FuncGraph: BatchNorm.20 Args: [0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), [5]: AbstractTuple{element[0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue)}, [6]: AbstractTuple{element[0]: AbstractTensor(shape: (1, 32, 40, 99), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[1]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[2]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[3]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue), element[4]: AbstractTensor(shape: (32), element: AbstractScalar(Type: Float32, Value: AnyValue, Shape: NoShape), value_ptr: 0x18b1728ef90, value: AnyValue)}, Parent: { Args: }}, FuncGraph: 0x18b20bb0cb0/BatchNorm.20
3 根因分析
根据报错信息提示，这应该是数据的类型不符合计算。
4 解决方案
构造数据集类的时候定义了mindspore的tensor，换成np数组就可以了。