Function suggest

hyperopt/atpe.py:1577–1621

(new_ids, domain, trials, seed)

def suggest(new_ids, domain, trials, seed):
    optimizer = ATPEOptimizer()

    # Convert the PyLL domain back into a descriptive form of hyperparameter space
    hyperparameterConfig = Hyperparameter.createHyperparameterConfigForHyperoptDomain(
        domain
    )

    results = optimizer.convertTrialsToResults(hyperparameterConfig, trials)

    # If there is a loss value that is negative, then we must increment the values so
    # they are all positive. This is because ATPE has been optimized only for positive
    # loss value
    if len(results) > 0:
        minVal = min(
            [result["loss"] for result in results if result["loss"] is not None]
        )
        if minVal < 0:
            for result in results:
                if result["loss"] is not None:
                    result["loss"] = result["loss"] - minVal + 0.1

    hyperparameters = Hyperparameter(hyperparameterConfig)

    rval = []
    for new_id in new_ids:
        parameters = optimizer.recommendNextParameters(
            hyperparameterConfig, results, currentTrials=[]
        )
        flatParameters = hyperparameters.convertToFlatValues(parameters)

        rval_results = [domain.new_result()]
        rval_miscs = [
            dict(
                tid=new_id,
                cmd=domain.cmd,
                workdir=domain.workdir,
                idxs={key: [0] for key in flatParameters},
                vals={key: [flatParameters[key]] for key in flatParameters},
            )
        ]

        rval.extend(trials.new_trial_docs([new_id], [None], rval_results, rval_miscs))

    return rval