Feature "image" is accessed differently if dataset is DCOR resource

[B-Hartmann]

System

Windows 10 Python 3.9 dclab 0.43.1

Minimal working example:

>>import dclab
>>ds_online = dclab.new_dataset("e4d59480-fa5b-c34e-0001-46a944afc8ea")
>>ds_offline = dclab.new_dataset("path/to/file.rtdc")

>>ds_online["image"].shape
(4544, 80, 250)

>>ds_offline["image"].shape
(1561, 80, 250)

>>ds_offline["image"][0:2,:,:]
array([[...]])

>>ds_online["image"][0:2,:,:]
Traceback (most recent call last):
  File "C:\Users\bhartma\.environments\env\lib\site-packages\IPython\core\interactiveshell.py", line 3457, in run_code
    exec(code_obj, self.user_global_ns, self.user_ns)
  File "<ipython-input-10-e2fa7e36cfcd>", line 1, in <module>
    ds_online["image"][0:2,:,:]
  File "C:\Users\bhartma\.environments\env\lib\site-packages\dclab\rtdc_dataset\fmt_dcor\features.py", line 26, in __getitem__
    indices = np.arange(len(self))[event]
IndexError: too many indices for array: array is 1-dimensional, but 3 were indexed

What I expected

So the shape of the data is the same, basically a 3 dimensional array, right? But then using slicing to access the first few images fails when accessing the dataset on DCOR.
I would have expected that this works, no matter if the dataset resource if online or offline.

[paulmueller]

Slicing non-scalar features from DCOR is not supported right now, simply because it is so inefficient (image data is read and converted to json by the server, then converted back to an array by dclab). If you wanted to work with non-scalar features, you should download the entire dataset.

I don't see a particular use case where slicing image data justifies not downloading the entire dataset. But I am open to discussing it.

[B-Hartmann]

Okay. For me, it works now just using [0:10] instead of [0:10, :, :], so that is fine.

The performance issue is reasonable! I don't have a better solution for this right now, I guess I will close the issue for now.

[paulmueller]

Ok thanks!