Lets say the user wants to run NUTS from pymc (call it pymc_nuts) on the posterior eight_schools_centered. However, the posterior database only has model code for Stan.
Currently we would get the error when running the inference. I think it would be good that we could check if an inference task is valid without having to run the inference(*). This would require inference methods to specify which modelling framework they work with. So essentially the inference method pymc_nuts would have "model_framework": "pymc" in addition to "language": "python" (from #6)
(*): This could be handy for example when running 1000 long-running experiments in a huge cluster as it would be good to catch as many errors as possible in the validation stage
eerolinna
commented
4 years ago
Lets say the user wants to run
NUTSfrompymc(call itpymc_nuts) on the posterioreight_schools_centered. However, the posterior database only has model code for Stan.Currently we would get the error when running the inference. I think it would be good that we could check if an inference task is valid without having to run the inference(*). This would require inference methods to specify which modelling framework they work with. So essentially the inference method
pymc_nutswould have"model_framework": "pymc"in addition to"language": "python"(from #6)(*): This could be handy for example when running 1000 long-running experiments in a huge cluster as it would be good to catch as many errors as possible in the validation stage