bruce-edelman / markpy Goto Github PK

Samples seem to converge to analytical values we generated data off of and show independence but the 90% confidence intervals seem off when we try to plot the signal.

Does not work properly for the 3d or 6d versions in test.py

Need to try some other algorithm besides Metroolis-Hastings.
Most framework should already be in place, just need to program the correct stepper objects in steppers.py

This is a bug with the Parallel Mark Chain object when it is recieveing the samples from each chain. we put it in an already initialized array of size nsteps but when we take the burn or ind samps we have a smaller length array. Need a way to fix this.

Started to fix this in the most recent commit (check the run fct of ParallelMark Chain in master branch)

It has not been tested yet so it still needs that

Figure out what we need to do to add in MPI support so we can run the Parralel MarkChain with nchains=ncores in your CPU to maximize performance

Need to add property to our MarkChain object and possible the Model/likliehoods for the Nwalkers

This will be the number of independent chains we use. Fix other features that requiring averaging / looking at each of the new chains.

Lastly we will need to fix anything we see slightly slowing it down since once we add this feature the computational time will increase drastically for testing