--check-convergence & --force-convergence

[QianXiaobo]

Hi Xin,

Thanks for your nice work. I am confused about following two parameters in InferDM model. I think they both mean stop optimization runs when convergence.

--check-convergence CHECK_CONVERGENCE
                        Start checking for convergence after a chosen number of optimizations. Stop optimization runs when convergence criteria are reached. BestFit results file
                        will be call <output_prefix>.InferDM.bestfits. Convergence not checked by default.
--force-convergence FORCE_CONVERGENCE
                        Start checking for convergence after a chosen number of optimizations. Only stop optimization once convergence criteria is reached. BestFit results file
                        will be call <output_prefix>.InferDM.bestfits. Convergence not checked by default.

I also read the user guide but still don't know the difference between them. Could you please give me more details or example?

Bests,
Xiaobo

[xin-huang]

Hello @QianXiaobo

Thanks for testing dadi-cli.

I believe the --check-convergence option will stop the program after the number of iterations specified by the --optimization argument, regardless of convergence. If it achieves converged results before reaching the specified number of iterations, it may stop early. However, the --force-convergence option will ensure the program continues running until it achieves the converged results.

Maybe @tjstruck could provide more details?

[tjstruck]

Hi @QianXiaobo, @xin-huang is correct.

--force-convergence will run dadi-cli until convergence is met.
--check-convergence will run dadi-cli until the number of optimizations requested through --optimization is completed and stop early if convergence is met.

I'll update the documentation to help make the distinction clearer.