Question about the CorrDiff Example

[MyGitHub-G]

when i run python train.py, the error:

Error executing job with overrides: []
Traceback (most recent call last):
  File "/public/home/shizhm2050/gexl/modulus/examples/generative/corrdiff/train.py", line 104, in main
    img_out_channels = len(dataset.output_channels())
  File "/public/home/shizhm2050/gexl/modulus/examples/generative/corrdiff/datasets/cwb.py", line 448, in output_channels
    return [out_channels[i] for i in self.out_channels]
  File "/public/home/shizhm2050/gexl/modulus/examples/generative/corrdiff/datasets/cwb.py", line 448, in <listcomp>
    return [out_channels[i] for i in self.out_channels]
IndexError: list index out of range

what is the reason?

and

cwb_train.yaml文件中in_channels: [0, 1, 2, 3, 4, 9, 10, 11, 12, 17, 18, 19]是什么意思，为什么不是0~19，代表20个变量呢

[MyGitHub-G]

cwb.py ZarrDataset class:

def input_channels(self):
        """Metadata for the input channels. A list of dictionaries, one for each channel"""
        in_channels = self._dataset.input_channels()
        # print(in_channels)  # 20个输入变量
        # print(self.in_channels) # [0, 1, 2, 3, 4, 9, 10, 11, 12, 17, 18, 19]
        in_channels = [in_channels[i] for i in self.in_channels]
        return in_channels

    def output_channels(self):
        """Metadata for the output channels. A list of dictionaries, one for each channel"""
        out_channels = self._dataset.output_channels()
        print(out_channels) 
        # [ChannelMetadata(name='maximum_radar_reflectivity', level='', auxiliary=False), 
        #   ChannelMetadata(name='temperature_2m', level='', auxiliary=False), 
        #   ChannelMetadata(name='eastward_wind_10m', level='', auxiliary=False), 
        #   ChannelMetadata(name='northward_wind_10m', level='', auxiliary=False)]
        # print(self.out_channels) #[0, 17, 18, 19]
        # return [out_channels[i] for i in self.out_channels]
        return [out_channels[i] for i in range(len(self.out_channels))]

in the output_channels function, the self.out_channels is [0, 17, 18, 19], not [0,1,2,3]
17 results the IndexError
i changed thereturn [out_channels[i] for i in self.out_channels]to return [out_channels[i] for i in range(len(self.out_channels))]
is it a bug?

[MyGitHub-G]

cwb.py ZarrDataset class getitem function:

# channels
        input = input[self.in_channels, :, :]
        # target = target[self.out_channels, :, :]
        target = target[range(len(self.out_channels)), :, :]

target = target[self.out_channels, :, :] results error IndexError: index 17 is out of bounds for axis 0 with size 4
changed to target = target[range(len(self.out_channels)), :, :]
is it a bug?

[zomosky]

cwb.py ZarrDataset class getitem function:

# channels
        input = input[self.in_channels, :, :]
        # target = target[self.out_channels, :, :]
        target = target[range(len(self.out_channels)), :, :]

target = target[self.out_channels, :, :] results error IndexError: index 17 is out of bounds for axis 0 with size 4 changed to target = target[range(len(self.out_channels)), :, :] is it a bug?

I have changed it in the conf/yaml by cfg.dataset.out_channels = [0, 1, 2, 3], also the channels means the channels choice in the zarr, you can see channels by zarr.tree()

我也觉得这是一个bug，也有可能是之前的训练集的结构不一样导致的，可以直接在yaml或者运行的时候改变channel就行，因为作为output的输出的数据集size为4，所以可以设置为cfg.dataset.out_channels = [0, 1, 2, 3] 可以通过简单的zarr库和zarr.tree()方法查看zarr数据库的结构

import zarr as zr
zrpath = '/home/cwa_dataset/cwa_dataset.zarr'
ze = zr.open(zrpath, mode = 'r')
zr.load(zrpath)
group = zr.open_consolidated(zrpath)
ze = zr.open(zrpath, mode = 'r')
zr.tree(ze)

[MyGitHub-G]

cwb.py ZarrDataset class getitem function:

# channels
        input = input[self.in_channels, :, :]
        # target = target[self.out_channels, :, :]
        target = target[range(len(self.out_channels)), :, :]

target = target[self.out_channels, :, :] results error IndexError: index 17 is out of bounds for axis 0 with size 4 changed to target = target[range(len(self.out_channels)), :, :] is it a bug?

I have changed it in the conf/yaml by cfg.dataset.out_channels = [0, 1, 2, 3], also the channels means the channels choice in the zarr, you can see channels by zarr.tree()

我也觉得这是一个bug，也有可能是之前的训练集的结构不一样导致的，可以直接在yaml或者运行的时候改变channel就行，因为作为output的输出的数据集size为4，所以可以设置为cfg.dataset.out_channels = [0, 1, 2, 3] 可以通过简单的zarr库和zarr.tree()方法查看zarr数据库的结构

import zarr as zr
zrpath = '/home/cwa_dataset/cwa_dataset.zarr'
ze = zr.open(zrpath, mode = 'r')
zr.load(zrpath)
group = zr.open_consolidated(zrpath)
ze = zr.open(zrpath, mode = 'r')
zr.tree(ze)