SI_LULC_pipeline move away from increasing integer sequence? about eo-learn HOT 2 CLOSED

Thanks for the suggestions.

We have also noticed that the filename has been changed. You can get the file on http://rkg.gov.si/GERK selecting the Grafični podatki RABA za celo Slovenijo (shape.rar ~ 500 MB) KoordSistem: D96/TM file. We'll update the notebooks accordingly.

Which parts of this workflow wouldn't work with your given labels?

Any more details would help us improve the example.

from eo-learn.

It's a suggestion and not a bug. I was just interested in doing this in my area. All works fine on your examples. :)

I tried to find an area that is suitably representative of the other patches. It's in the inner City of Glasgow. CRS.UTM_30N

[27]:

class_names = [entry.class_name for entry in LULC]


print('             Class              =  F1  | Recall | Precision')
print('         --------------------------------------------------')
for idx, lulctype in enumerate([class_names[idx] for idx in class_labels]):
    print('         * {0:20s} = {1:2.1f} |  {2:2.1f}  | {3:2.1f}'.format(lulctype,
                                                                         f1_scores[idx] * 100,
                                                                         recall[idx] * 100,
                                                                         precision[idx] * 100))

I wrote my own adhoc code.

I couldn't test if all training labels where in the other patches as the patch size is quite small.

Labels and plabels then don't mix. So the confusion matrix goes AWOL showing some empty rows.

[29]:
classes=[name for idx, name in enumerate(class_names) if idx in class_labels]

I just thought it might save others some bother if they try to play around with their own neck of woods. :)

from eo-learn.