Interpret data to normalize as ndarrays

[ngoldbaum]

This follows on from #6622. That PR fixed the case where we were calling imshow with linear normalization. This fixes the case when we call imshow with log, symlog, or power normalization. For example:

import numpy as np

from matplotlib import pyplot as plt
from matplotlib.colors import LogNorm
from yt.units import km

plt.imshow(np.random.random((16, 16))*km, norm=LogNorm())

Currently this dies with a unit error in LogNorm. Analogous to the fix applied in #6622 the fix is to interpret data handed to us by a user as an ndarray.

[efiring]

Can this as well as the #6622 fix be consolidated into Normalize.process_value()? That seems to me to be the logical place to strip off the units.

[ngoldbaum]

Added logic to process_value. Unfortunately I couldn't use the new return value I added to process_value for all the normalizers since some of them do some additional data processing after the call to process_value.

I also added a test.

[efiring]

A simple one-line change replaces everything else. I think that means #6622 can be reverted as well.

[efiring]

I think that all you really need to do here is add the kwarg subok=False to the np.ma.array() line returning result. Then result will be a plain masked array, and result.data will be a plain ndarray, so you will not need to extract and return it here. The reason all this came up is that np.ma.array has a surprising default of subok=True, unlike np.array, which has the default subok=False.

[ngoldbaum]

@efiring subok didn't seem to work (I think it only affects the type of result, not result.data) but the approach I just pushed does seem to work. Are you ok with just unconditionally calling np.asarray unless we're passed a masked array?

[efiring]

Try this substitution:

# result = np.ma.array(value, dtype=dtype, copy=True)
mask = np.ma.getmask(value)
data = np.asarray(np.ma.getdata(value))
result = np.ma.array(data, mask=mask, dtype=dtype, copy=True)

[ngoldbaum]

Cool that works. Thanks for the suggestions, if it isn't clear I know basically nothing about masked arrays :)