Parallel support for bioscrape inference

[ayush9pandey]

Changes:

• Remove the deprecated threads argument passed in emcee and add parallel support with multiprocessing.

• Fix inference tests to account for inference changes

• Add two emcee benchmarks: series and paralle

• Add example for running a biocircuit model inference with parallel support

One additional (unrelated but important) minor change: specify versions for scipy and numpy in requirements and pyproject. Numpy version greater than 2.0 breaks the installation (to be fixed later).

[WilliamIX]

Looks good! Did you try this and was it any faster? In the past I tried some amount of parallelization and found that on Linux boxes at least, Pool slowed things down (I think because numpy operations were being done in parallel so the OS started fighting for threads).

[ayush9pandey]

Yes, the script emcee_parallel.py runs ~10x faster than emcee_series.py on my laptop CPU with 20 cores. The key is that this is an artificially slowed down likelihood function where it 'sleeps' for a few seconds in each iteration. For simple likelihood functions, the overhead with setting up parallel ends up slowing it down, as you suggested. So, parallelization only helps for complex likelihoods.

For the real example parallel_bioscrape_inference.py with 1 Hill function and identifying 3 params, I saw a 3x improvement on the HPC.