changed robustness function to first draw normal distributions and then add according to weights instead other way around. if first blow up means and stds and then resample cdfs too smooth which leads to artifical increase in R

added helper scripts to create configs for all time periods, added additional plotting scripts under version control

added removal of best and second best model for R calculation, needs revision when using weights for performance AND independence

rpsc (A2) is calculated from multi-model mean historical and future projections (not individual model runs), do not calculate rpss2 for efficiency (needs sorted cv2all, fvall etc.)

updated robustness_function_gufunc according to new version of robustness_function, simplified, use cdf

use cdf calculation instead interpolation, repetitions for all models at once, add break in loop if std over R is small

added saving of global mean change in multi_model_robustness.py, do not load numba in robustness function, not used at the moment

added season in plot name in postproc/plot_cmip6_cmip5_cmip3.ncl, only two outputs from multitests in multi_model_robustness.py, fixed weird error from symbols? (deleted line break) in functions/robustness_function.py

added weighting option for robustness_function. use robustness_function at every grid point with apply_ufunc and vectorize=True. tried using dask and numba, issue with R being dask array, something wrong

use xr.apply_ufunc to apply robustness_function at every grid point, even if weighted apply significance test on raw data

simplify ncl plotting, use hatching saved from python code for consistence,  use doubles where double saved in netcdfs

updated .gitignore to not track *.pyc files, undo change in plotting.py for hatching 'xx', does not work, looks the same as 'x'

changes for CMIP6 in main, added project to distinguish between CMIPs, for CMIP6 need to check that matching historical file is available, changed reading of weights to also read filenames from weights file to ensure identical ensemble. added postproc/plot_cmip6_cmip5_cmip3.ncl under version control, no content changes in functions/robustness_function.py

changes needed for new CMIP6-ng usage in multi_model_robustness.py, updated README with more info, added files calc_performance.py and postproc/plot_rmse_vs_R.py under version control

added option for CMIP3,5,6, changes needed because of slightly different data structctures, added environment info in README

implementation of weights for robustness (1000, include models as many times as needed by their weight), alway plot as contourf for consistency (otherwise looks strange), make sure correct fill value (not NaN, cannot be read by ncl and causes problems plotting) when writing data to netcdfs, cleaned up code