Migrating this issue as it's probably better here than on the DSToolkit.
Our current code for MCMC-based model evaluation uses emcee v2 and the LMFit interface for convenience. It also includes options for using emcee's parallel tempering algorithm (also via the LMFit interface).
Our existing Python code for Mobius requires emcee=2.2.1 and lmfit=0.9.14 (these versions are currently pinned on the JupyterHub/DSToolkit for convenience).
In v3, emcee introduced substantial changes to the API, which include removing the PTSampler completely (it's now a separate project here, but maintenance seems patchy). LMFit has also reduced/removed support for emcee, which is a shame.
At some point we need to refactor our Python code to use the latest version of emcee (and probably remove any dependencies on LMFit).
Migrating this issue as it's probably better here than on the DSToolkit.
Our current code for MCMC-based model evaluation uses emcee v2 and the LMFit interface for convenience. It also includes options for using emcee's parallel tempering algorithm (also via the LMFit interface).
Our existing Python code for Mobius requires
emcee=2.2.1andlmfit=0.9.14(these versions are currently pinned on the JupyterHub/DSToolkit for convenience).In v3, emcee introduced substantial changes to the API, which include removing the PTSampler completely (it's now a separate project here, but maintenance seems patchy). LMFit has also reduced/removed support for emcee, which is a shame.
At some point we need to refactor our Python code to use the latest version of emcee (and probably remove any dependencies on LMFit).