`reclassify_raster` does not handle `numpy.nan` nodata

[davemfish]

It's somewhat common for nan to be used as a raster nodata value. reclassify_raster can either reclassify nodata pixels -- if it is listed in the value map -- or propagate them as nodata in the output. Neither of these will work for nan because we use numpy.isclose() without passing equal_nan.

I feel like we've discussed this in the past, but I don't remember if there's a reason we did not address it. It came up in #419 . @claire-simpson do you remember?

[claire-simpson]

I think we could use equal_nan! I believe that the issue I noted in #419:

a numpy.nan key will cause an issue later in the function

is related to these 2 lines:

keys = sorted(numpy.array(list(value_map_copy.keys())))
values = numpy.array([value_map_copy[x] for x in keys])

There is some weirdness when trying to sort a numpy.nan key (in value_map_copy) in this line: keys = sorted(numpy.array(list(value_map_copy.keys()))) that ends up causing a KeyError when trying to create the values array.I think this is because the dictionary key is initially numpy.nan but then gets converted to numpy.float64(nan) which is different and therefore not a valid key in value_map_copy.

e.g.

value_map_copy = {1:11.0, 3:33, numpy.nan: -999}
print(value_map_copy) # {1: 11.0, 3: 33, nan: -999}
keys = sorted(numpy.array(list(value_map_copy.keys())))
print(keys) # [np.float64(1.0), np.float64(3.0), np.float64(nan)]
values = numpy.array([value_map_copy[x] for x in keys]) # raises KeyError: np.float64(nan)

However I think using equal_nan could actually solve this issue if the input raster's nodata value is numpy.nan, because, if nan is initially in the value_map, it would get deleted before values = numpy.array([value_map_copy[x] for x in keys]). Or maybe we should just be detecting nans anyway and handling them separately. TLDR; Using equal_nan seems conceptually correct to me and we should handle nan values in the value_map so they don't cause a KeyError.

[davemfish]

This makes sense; thanks for the refresher.

[phargogh]

This was resolved in #465