Inference

Related options

• Base options:

  • Global optimizer - name of global optimization (Genetic algorithm).

  • Local optimizer - name of local search algorithm (BFGS_log).

• Additional options:

  • Global maxiter - number of iterations for global optimization (None).

  • Global maxeval - number of log-likelihood evaluations for global optimization (None).

  • Global log transform indicates if data is log-transformed in global optimization (False).

  • Local maxiter - number of iterations for local search algorithm (None).

  • Local maxeval - number of log-likelihood evaluations for local search algorithm (None).

  • Local log transform indicates if data is log-transformed in local search algorithm (True).

  • Num init const - scaling constant (to number of model parameters) to define number of random points in initial design (10).

GADMA could be customized to use different to default optimization algorithms:

• List of all available global optimizers.

• List of all available local optimizers.

More specific information:

• How to choose optimizer

• Demographic inference for more than three populations

GADMA’s inference could be run both from command-line and from Python directly.

Command-line

Usage of GADMA:

$ gadma --help

GADMA version 2.0.0 by Ekaterina Noskova (ekaterina.e.noskova@gmail.com)
Usage:
            gadma   -p/--params <params_file>
                    -e/--extra <extra_params_file>


Instead/With -p/--params and -e/--extra option you can set:
            -o/--output <output_dir>        output directory.
            -i/--input <in.fs>/<in.txt>     input file with AFS or in dadi format.
            --resume <resume_dir>           resume another launch from <resume_dir>.
            --only_models                   flag to take models only from another launch (--resume option).

            -h/--help               show this help message and exit.
            -v/--version            show version and exit.
            --test                  run test case.

    In case of any questions or problems, please contact: ekaterina.e.noskova@gmail.com

Resume launch

Related options

• Base options:

  • Resume from

• Additional options:

  • Only models

To resume interrupted launch one can use --resume command-line option or set Resume from in the parameter file. One needs to set the output directory of the previous run.

If neither Output directory or -o/--output is not specified, GADMA will continue evaluation in the directory: <previous_output_dir>_resumed.

Only models

GADMA can resume launch taking final models only from the previous run. This means, that it is not the usual resumption, but run from some initial values. It is useful, for example, when one has to run GADMA with some small grid size for dadi and then wants to restart it with a greater number of grid points. To do so, one should set the command-line option --only_models with --resume or specify Only models option in the parameter file to True.

Run optimizer from Python

Genetic algorithm pipeline from GAGMA is available from gadma’s API by calling gadma.Inference.optimize_ga function. It is like usual optimization functions in dadi and moments. The short notice from the API: