Resume Optimization

It is possible to resume a previous optimization run in HyperMapper. This is useful to either increase the number of optimization iterations performed or to resume an optimization run that ended prematurely. This feature is activated in the json scenario file by triggering the resume optimization flag and providing the output csv file generated by HyperMapper. For example:

{
    "application_name": "branin",
    "optimization_objectives": ["Value"],
    "optimization_iterations": 20,
    "resume_optimization": true,
    "resume_optimization_data": "branin_resume_samples.csv",
    "input_parameters" : {
        "x1": {
            "parameter_type" : "real",
            "values" : [-5, 10]
        },
        "x2": {
            "parameter_type" : "real",
            "values" : [0, 15]
        }
    }
}

Users can provide the complete filepath to the resume_data_file (e.g. /home/user/branin_resume_samples.csv) or the path from HyperMapper's home directory. When resuming a previous optimization run, HyperMapper will notify the user and inform how many samples it read from the resume_data_file:

Resumed optimization, number of samples = 16 ......

Based on the number of samples already explored, HyperMapper will automatically determine whether it should resume from the design of experiment phase or the optimization phase.

Note. The resume optimization feature is only available for the bayesian_optimization method for now.