Coordinates attributes disappear when using m.predict(inplace=True)

[AndreaGarciaJuan]

I am coding the automatic detection of coordinates using coordinates attribute 'axis'. It is working well when I use the dataset before labels prediction (you can see atributes in the screen shot below):

But when I use predict function with the option inplace=True, coordinates attributes disappear from dataset (general attributes and variables attributes remain in dataset):

Do you know why is this happening?

[gmaze]

oh, that should not happens ! so, pyxpcm seems to delete all attributes of coordinates, but not variables do you have the same issue with another dataset ?

[AndreaGarciaJuan]

I have the same problem with ISAS and CMCC reanalysis

[AndreaGarciaJuan]

And when I do ds['depth'] = -np.abs(ds['depth'].values) depth attributes also disapear. I am trying to fix it.

[gmaze]

And when I do ds['depth'] = -np.abs(ds['depth'].values) depth attributes also disapear. I am trying to fix it.

Try:

ds['depth'].values = -np.abs(ds['depth'].values)

[gmaze]

I have the same problem with ISAS and CMCC reanalysis

ok, these are gridded products, what about argo dataset ?

[AndreaGarciaJuan]

I have the same problem with ISAS and CMCC reanalysis

ok, these are gridded products, what about argo dataset ?

For argo data I am using argopy. After doing ds = argo_loader.region([-6, 35, 30, 46, 0, 1000, '2012', '2013']).to_xarray() I can find coordinates attributes but after doing dsp = ds.argo.point2profile() they disappear. So there are no coordinates attributes when I apply .predict

[gmaze]

Ok, it's coming from the xarray accessor. I'll check deeper into this, in the mean time try to work with inplace=False

[AndreaGarciaJuan]

Ok! Thanks!

[gmaze]

@AndreaGarciaJuan can you please post here the python code to reproduce this error with each of the datasets ?

[AndreaGarciaJuan]

For ISAS

import xarray as xr
import numpy as np

import pyxpcm
from pyxpcm.models import pcm

# ISAS dataset
file_path = '/home5/pharos/REFERENCE_DATA/OCEAN_REP/ISAS/ISAS15/ANA/ISAS_DT/ARGO_ONLY/2015/'

#time range
start_date = 20150115
end_date = 20150115
time_interval = 100 # TODO: easier way to get time interval, date type

#spatial extent
lon_extent = [-6,35]
lat_extent = [30,45]

# get list of paths
dates_range = np.arange(start_date, end_date + time_interval, time_interval)

ds_paths = []
for dt in dates_range:
    ds_paths.append([file_path + 'ISAS15_ARGO_' + str(dt) +'_fld_TEMP.nc'])

#open all data files
ds = xr.open_mfdataset(ds_paths,combine='by_coords', concat_dim='time')

ds = ds.sel(latitude=slice(lat_extent[0],lat_extent[1]), longitude=slice(lon_extent[0], lon_extent[1])) 
# !!!!!! after changing depth values, attributes disappear (ds['depth'].values = -np.abs(ds['depth'].values) is not working)
ds['depth'] = -np.abs(ds['depth'].values)

#Choise of z and pcm features (very important to make a good choise) 
z = np.arange(0.,-2000,-10.)
pcm_features = {'temperature': z}
K=6
m = pcm(K=K, features=pcm_features)

# fit
features_in_ds = {'temperature': 'TEMP'}
features_zdim='depth'
m.fit(ds, features=features_in_ds, dim=features_zdim)

#classify data
m.predict(ds, features=features_in_ds, dim=features_zdim, inplace=True);

# !!!!!! after predict no attributes in coordinates
ds.longitude.attrs

[AndreaGarciaJuan]

For Argo data

import xarray as xr
import numpy as np

import argopy
from argopy import DataFetcher as ArgoDataFetcher
argo_loader = ArgoDataFetcher()

# get data
ds = argo_loader.region([-6, 35, 30, 46, 0, 1000, '2012', '2013']).to_xarray()

dsp = ds.argo.point2profile()
# !!!!!! after point2profile no attributes in coordinates
dsp.LONGITUDE.attrs

[gmaze]

For Argo data

Was due to argopy internals, this is fixed for argopy.

[gmaze]

also reported here: https://github.com/obidam/pyxpcm/issues/30 work in progress at: https://github.com/obidam/pyxpcm/pull/31