Closed lilianschuster closed 2 years ago
added notebook with options to emulate a MB model --> https://github.com/OGGM/massbalance-sandbox/docs/emulator_bayes_calib/bayesian_gaussian_process_emulator_tests_2022_09.ipynb
GPyTorch using exact GP -> same as what is done in PyGEM, see also https://docs.gpytorch.ai/en/stable/examples/01_Exact_GPs/Simple_GP_Regression.html)
In addition, started to look at the Bayesian GP of GPyTorch but did not manage to actually include the priors and geodetic observation data into the emulator (https://docs.gpytorch.ai/en/latest/examples/01_Exact_GPs/GP_Regression_Fully_Bayesian.html)
Also looked at Gaussion Process Regression using PyMC3, however I still have the old problem that I don't know how to merge emulator to the geodetic calibration part ... e.g. see also (https://discourse.pymc.io/t/use-exact-gaussian-process-model-from-gpytorch-as-emulator-in-pymc3/8680)
for the emulators, I use a test dataset from the HEF glacier (table of temp. bias, prcp. fac , melt_f with resulting mean 2000-2020 specific MB values, inside of same folder as the notebook)
more infos inside of the notebook
added notebook with options to emulate a MB model --> https://github.com/OGGM/massbalance-sandbox/docs/emulator_bayes_calib/bayesian_gaussian_process_emulator_tests_2022_09.ipynb
GPyTorch using exact GP -> same as what is done in PyGEM, see also https://docs.gpytorch.ai/en/stable/examples/01_Exact_GPs/Simple_GP_Regression.html)
In addition, started to look at the Bayesian GP of GPyTorch but did not manage to actually include the priors and geodetic observation data into the emulator (https://docs.gpytorch.ai/en/latest/examples/01_Exact_GPs/GP_Regression_Fully_Bayesian.html)
Also looked at Gaussion Process Regression using PyMC3, however I still have the old problem that I don't know how to merge emulator to the geodetic calibration part ... e.g. see also (https://discourse.pymc.io/t/use-exact-gaussian-process-model-from-gpytorch-as-emulator-in-pymc3/8680)
for the emulators, I use a test dataset from the HEF glacier (table of temp. bias, prcp. fac , melt_f with resulting mean 2000-2020 specific MB values, inside of same folder as the notebook)
more infos inside of the notebook