Describe the bug

when a path is given to load data, and that data is incorrectly formatted, load_classification deletes the file and directory. It should not do this!

Steps/Code to reproduce the bug

from aeon.datasets import load_classification
X, y = load_classification("GunPoint", extract_path="C:\\Temp\\")

this works and loads the data as expected from directory C:\Temp\GunPoint. However, if a dataset is incorrectly formatted, it deletes the directory.

from aeon.datasets import load_classification
X, y = load_classification("NBack_combined", extract_path="C:\\Temp\\")

raises this error after deleting the directory
File "C:\Code\aeon\aeon\datasets_data_loaders.py", line 1382, in load_classification
raise ValueError(error_str)
ValueError: Invalid dataset name =Test that is not available on extract path =C:\Temp. Nor is it available on https://timeseriesclassification.com/ or zenodo.

For this file, the error is having one channel a different length to the others, but Ive not tested it with other errors. This

from aeon.datasets import load_classification, load_from_ts_file
X, y = load_from_ts_file("C:\\Temp\\NBack_combined")
print(X.shape)

Expected results

load_classification behaves the same as load_from_ts_file

Actual results

throws the (correct) error

 File "C:\Code\aeon\aeon\datasets\_data_loaders.py", line 282, in load_from_ts_file
    data, y, meta_data = _load_data(file, meta_data)
  File "C:\Code\aeon\aeon\datasets\_data_loaders.py", line 215, in _load_data
    raise OSError(
OSError: channel 0 in case 11 has a different number of observations to the other channels. Saw 135 in the first channel but 136 in the channel 0. The meta data specifies equal length == True. But even if series length are unequal, all channels for a single case must be the same length

Versions

No response