Question about installation

[opierpaoli]

Hi,

I am trying to install and run some code using EOmaps, and while I was successfully able to do this on one computer, I am not able to replicate the same figures with a copy of the same conda environment on my other laptop. When I use the plot_map function I get this error: RuntimeWarning: invalid value encountered in subtract delta = stop - start which makes the figure extents out of bounds. I am using python 3.8.13, numpy 1.21.5, cartopy 0.20.2, and EOmaps 3.5. Any ideas on how to fix this or is there a .yml file that could be provided for me with the correct versions of the required packages?

Thanks!

[raphaelquast]

Hi,
sorry to hear that but from what you report here it's quite hard to know where your error is coming from...

• what do you mean by a "copy of the same conda environment"... did you export a .yml file and installed a new env from that?
• can you provide the code or at least the full error traceback message?

---

In general, the fastest way to get EOmaps running is the following:

• install mamba to speed up the process
  conda install -c conda-forge mamba
• create a new env with eomaps (and matplotlib to get all the backends)
  mamba create -n NAME -c conda-forge eomaps matplotlib
  (NAME is your env name and the python-version can be anything above 3.7)

[opierpaoli]

Hi!
To answer: what do you mean by a "copy of the same conda environment"... did you export a .yml file and installed a new env from that?
Yes, I tried this, but it didn't work. But I've also used the same commands on 2 different computers and one environment
works and the other has a numpy core error.

Here is the full error message:
\path\to\env\site-packages\numpy\core\function_base.py:134: RuntimeWarning: invalid value encountered in subtract delta = stop - start

this error appears once I use the plot_map function. I can still draw coastlines and other features.

I tried installing using :

conda create -n NAME -c conda-forge eomaps matplotlib

and I still get the same issue.

[opierpaoli]

just tried with mamba as well, and I'm still getting the same error. Specifically when using plot_map.

[raphaelquast]

hey, ok so that seems like a quite strange problem...

it certainly originates from some call to np.linspace:
https://github.com/numpy/numpy/blob/0eaa40db3a2f6bef26c0d7cc28eacf4f6d461917/numpy/core/function_base.py#L134

However, since I don't have any clue to what's happening yet, I can only give a few more suggestions on how we could proceed to find out what's going on:

• concerning the error-message... this is not an error, this is a warning... in theory it should not break the code...
  The only way I can get the same warning is with something like: np.array([np.inf, 1, 2,3]) - np.array([np.inf, 1,2,3])
  (e.g. only undefined forms such as np.inf - np.inf give such a warning)
  From this I have some basic suggestions that might solve this
  (however none of them explain why it happens only on one computer and not on the other):

  • are you sure your input-data does not contain points at infinity (e.g. np.inf) ?
  • are you sure you provide the correct crs for the coordinates?
    (m.set_data(data, xcoord, ycoord, crs=???) otherwise reprojections might result in np.inf)

• can you provide a minimal reproducible example that I could try on my side?

• what operating system are you working on? (is it the same on both computers?)

• did you try with a fresh conda? (maybe the issue is originating from the conda-installation...)

  • you don't need to uninstall your existing setup... just use miniconda
    to install another conda-instance and then try this again:

    conda install -c conda-forge mamba
    mamba create -n Name -c conda-forge eomaps matpltolib
    

[raphaelquast]

While this is of course possible, I've never heard of a windows-OS version causing such a problem....
After all the conda-installation should be very much self-contained and the only OS-dependency that
I can think of are the precompiled C-extensions of libraries such as numpy...
(which might be resolved with a completely fresh miniconda env as suggested above)

Unfortunately I'm also on 19044 so I can't reproduce your errors on my side...

Can you provide some minimal set of code that causes this error ?
For example does this already result in a problem:

from eomaps import Maps
import numpy as np
lon, lat = np.meshgrid(*np.linspace([-50,-30], [50, 30], 100).T)

m = Maps()
m.set_data(lon, lon, lat)
m.plot_map()

.... and just out of curiosity (since you mentioned setting the plot-extent)... can you also try with:

m.plot_map(set_extent=False)

[opierpaoli]

I was using a fresh install of Anaconda, but I could try miniconda as well. I am also able to use the code you just sent and everything seems to be working properly.

You were actually helping me out with code over on StackOverflow with this post: https://stackoverflow.com/questions/71804525/cartopy-coastlines-not-lining-up-with-imshow-projection/71956219#71956219

from eomaps import Maps
import numpy as np
from pyhdf.SD import SD, SDC

data = SD("Copy of greenland_PM01_L2.2007061414150564.058140.1.46-public-release.hdf", SDC.READ)

#get the actual coordinates of your datapoints
lon, lat = data.select("subset_lon")[:], data.select("subset_lat")[:]
#mask nan-values
mask = np.logical_and(np.isfinite(lon), np.isfinite(lat))
lon, lat = lon[mask], lat[mask]

#get the colors you want to use
R = data.select("mband07")[:][mask]
G = data.select("mband02")[:][mask]
B = data.select("mband01")[:][mask]
RGB = list(zip(R, G, B)) # a list of rgb-tuples

#plot the data
m = Maps(Maps.CRS.Sinusoidal())
m.add_feature.preset.coastline()
m.set_shape.rectangles(radius=3100, radius_crs=Maps.CRS.Sinusoidal())
# use "dummy" values since we provide explicit colors
m.set_data(data=np.empty(lon.shape), xcoord=lon, ycoord=lat, crs=4326)
m.plot_map(set_extent=True, color=RGB)

I've since changed the code slightly to use a different projection. There's code that someone else sent using Basemap and I have the same issues where the set_extent works on one computer and not the other. With the computer that does work, the figures I am able to make with this library are amazing, so thank you again!

[raphaelquast]

aaah, OK I see, nice to hear that (at least on one PC) the library works nicely 😅
There is in fact a part in the code I've provided that might be OS-dependent.... np.empty(lon.shape) !
(it just initializes an empty array without caring about what's been allocated at the memory-location before...)
maybe try and replace it with something that is for sure not a very large number... such as np.full(lon.shape, 0)

[raphaelquast]

In the next release of EOmaps (v3.6 coming soon...) "coordinate-only" datasets can be plotted as well (e.g. using something like m.set_data(None, lon, lat)) but in v3.5 that's not yet possible

[opierpaoli]

I changed it and I get the same warning as before. I guess it has to be something with how the different os's behave with infinite numbers then?