Question

I have a subclassed numpy array – a “spectroscopic axis” with metadata such as units.

I can create a copy of any instance using arr.copy or a view using, e.g., arr2=arr[2:5].

Is there any way to create a new instance of one of these arrays with a different shape but keeping the same metadata / data type?

Answer 1

A simple solution would be to define a .reshape method in your subclass. This .reshape method should call the __array_finalize__ method that would be in charge of copying the metadata from your object.

For example, the MaskedArray subclass uses a _update_from private method that copies some metadata (_fill_value, _optinfo…) from one object to another or sets defaults for these metadata if they are not yet available. The _update_method is called within __array_finalize__. You could follow the sources as an example.