Multiple linear regression only returning coefficient for first independent variable

[loganlongacre]

hi there, I'm switching to pingouin from sklearn and I'm currently attempting to perform a multiple linear regression at each gridpoint using modeled weather data stored in netcdf files. in order to do this, I have a triple for loop structured like so:

data = [model1, model2, model3, model4, model5, model6, model7]

modelCoefs = np.zeros((len(ACCESS_CM2.lat), len(ACCESS_CM2.lon), len(paths), 4))

for i, j in enumerate(data):
     for k in len(data[i].lat):
           for l in len(data[i].lon):
                    dframe = pd.DataFrame({'EIS': data[i].isel(lat=k, lon=l).eis.values,
                                                              'HUR': data[i].isel(lat=k, lon=l, plev=3).hur.values,
                                                              'WAP': data[i].isel(lat=k, lon=l, plev=3).wap.values,
                                                              'SST': data[i].isel(lat=k, lon=l,).sst.values,
                                                              'LCF': data[i].isel(lat=k, lon=l).cf.values})
                    modelCoefs[k, l, i, :] = pg.linear_regression(dframe[['EIS','HUR','WAP','SST']], dframe['LCF'].coef) 

when I execute this code, the linear_regression command only returns regression coefficients for the intercept & EIS. is there a problem with how I've constructed the dataframe?

(for reference, the data are shaped like so: latitude x longitude x time x level. the latitude grid is np.linspace(-90,90,144), the longitude np.linspace(-180,180,192), time pd.daterange('20000101','20141231',freq='MS'), and level is a custom logarithmically decreasing range between 1000 hPa (surface) and 1 hPa (top of atmosphere) with 19 entries. EIS, SST, and LCF do not have level data; all 5 columns in the pd df are (should be!) a timeseries at a single point in 3d space).

[loganlongacre]

think I figured it out: columns with nan/infinitesimal values are problematic and I think there was an issue stemming from that wrt rank. changed preprocessing of nan from mean to 0 and that fixed it.