search

SysBioChalmers / BayesianGEM

This repository contains code and data needed to replicate the analysis carried out in the manuscript Li G, et al. 2020.
GNU General Public License v3.0
9 stars 4 forks source link

Failure to reach convergence #2

Open yaccos opened 3 years ago

yaccos commented 3 years ago

I have tried to replicate your results by running the script code/gem_smcabc_at_three_conditions.py as-is. Seven days after starting the computations on 48 virtual 2.4 GHz Intel Broadwell processors, the script was still running. Looking at the output of the script seems to indicate that the particles in the population were all the same and did not change over the course of the computations. For instance, I get identical r2-values no matter how long the script was run (taken from concurrent output):

Model r2 {'rae': -3.0077091095978403, 'chemostat': -1.036103450504834, 'ran': -3.0873882602816964}

Have you experienced such problems yourself or other issues with the convergence of the approximate Baysesian inference technique?

yaccos commented 3 years ago

Also, I see that you have not provided a random seed for the script. May this be of importance?

yaccos commented 3 years ago

I have now found a possible culpit of the problem. When changing the error handling code (see this branch https://github.com/AlmaasLab/BayesianGEM/tree/debug_reproducibility) it turns out that the code originally catched error which had nothing to do with the feasibility of the problem. In my case I got: 

gurobipy.GurobiError: Model too large for size-limited license; visit https://www.gurobi.com/free-trial for a full license

However, all errors which arose were treated as if the problem was infeasible without telling about the actual source of the problem. I will fix the issue with Gurobi and try again.