Keep the original dtype of the data when stretching #150
Conversation
|
This latest change to the tests made me think: do we allow integer data? If yes, should I check for that and then use |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #150 +/- ##
==========================================
+ Coverage 91.23% 91.28% +0.05%
==========================================
Files 11 11
Lines 3867 3890 +23
==========================================
+ Hits 3528 3551 +23
Misses 339 339
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Sentry. |
|
Integer data in any part of Satpy? Or a specific part? I expect all category products to be integers. And palletized images are also integers I think. |
|
Let me rephrase: are any integer data expected to be passed to |
|
I think integer data can be stretched. For example, suppose I have data that are discrete (categorical or not) and range from 0 to 10, that I want to write as a greyscale image in which the shades of grey can be seen. Then I'd want to stretch it to cover 0 to 255. In that case |
|
Thanks @gerritholl I'll see if @djhoese or @mraspaud have thoughs on this before starting anything. |
|
Restarting failed readthedocs job as it appeared to be a download SSL certificate error. |
There was a problem hiding this comment.
This as-is looks fine to me. I am concerned about the integer types that you mentioned.
I think we always want the result of stretching to be floating point though as trollimage does a 2 step process of scaling to the image format (data array -> normalized -> scale to file dtype). So maybe it is just easier to set these to 32-bit floats? Or do a check if the input array is a floating type and if not default to a 32-bit float for the scale factor.
| arr = np.arange(75).reshape(5, 5, 3).astype(float) | ||
| data = xr.DataArray(arr.copy(), dims=['y', 'x', 'bands'], | ||
| coords={'bands': ['R', 'G', 'B']}) | ||
| img = xrimage.XRImage(data) | ||
| img.crude_stretch(0, 74) | ||
| assert img.data.dtype == float |
There was a problem hiding this comment.
I'm not sure how important of a check this is unless it is switched to 32-bit float, but 🤷♂️ not a big deal
There was a problem hiding this comment.
This is there just to make sure everything is not changed to float32 ("keep the original dtype").
The stretching forces the dtype to
float64. This PR makes sure the original dtype is preserved.git diff origin/master **/*py | flake8 --diff(remove if you did not edit any Python files)