search

holoviz-topics / EarthML

Tools for working with machine learning in earth science
https://earthml.holoviz.org
BSD 3-Clause "New" or "Revised" License
94 stars 21 forks source link

Landsat multi-band spectral clustering #15

Closed mmccarty closed 5 years ago

mmccarty commented 5 years ago

@jbednar, I think I've got this working now for multiple bands. I've committed with a smaller sample to help with debugging the labels plot at the end.

@jlstevens Can you take a look at the labels plot at the end of the notebook. The artifacts we saw earlier seem to be a plotting issue. Note, the matplotlib version looks good.

screen shot 2018-09-07 at 2 27 14 pm

mmccarty commented 5 years ago

@jlstevens I think I've figured it out. The problem was regrid which I don't think we need. PR updated.

jlstevens commented 5 years ago

@mmccarty As you noticed, to don't have to use regrid but you should be able to. @jbednar Any idea why regrid might cause such artifacts? Seems very likely that something is funny with the datashader output...

jbednar commented 5 years ago

Right; regrid should be safe in general; it should just resample, and with a uniform pattern like on the lake area it should have very little effect. It would be nice to pull out a more specific test case that we can investigate.

mmccarty commented 5 years ago

@jbednar @jlstevens I'll trying to come up with a minimal example that reproduces the problem.