/// <summary>
/// Samples new hyperparameters for the trainer, and sets them.
/// Returns true if success (new hyperparameters were suggested and set). Else, returns false.
/// </summary>
private static bool SampleHyperparameters(MLContext context, SuggestedTrainer trainer, IEnumerable <SuggestedPipelineRunDetail> history, bool isMaximizingMetric)
{
var sps = ConvertToValueGenerators(trainer.SweepParams);
var sweeper = new SmacSweeper(context,
new SmacSweeper.Arguments
{
SweptParameters = sps
});
IEnumerable <SuggestedPipelineRunDetail> historyToUse = history
.Where(r => r.RunSucceeded && r.Pipeline.Trainer.TrainerName == trainer.TrainerName && r.Pipeline.Trainer.HyperParamSet != null && r.Pipeline.Trainer.HyperParamSet.Any());
// get new set of hyperparameter values
var proposedParamSet = sweeper.ProposeSweeps(1, historyToUse.Select(h => h.ToRunResult(isMaximizingMetric))).First();
if (!proposedParamSet.Any())
{
return(false);
}
// associate proposed parameter set with trainer, so that smart hyperparameter
// sweepers (like KDO) can map them back.
trainer.SetHyperparamValues(proposedParamSet);
return(true);
}
public SuggestedPipeline(IEnumerable <SuggestedTransform> transforms,
IEnumerable <SuggestedTransform> transformsPostTrainer,
SuggestedTrainer trainer,
MLContext context,
bool cacheBeforeTrainer)
{
Transforms = transforms.Select(t => t.Clone()).ToList();
TransformsPostTrainer = transformsPostTrainer.Select(t => t.Clone()).ToList();
Trainer = trainer.Clone();
_context = context;
_cacheBeforeTrainer = cacheBeforeTrainer;
}
public static SuggestedPipeline Build(MLContext context,
ICollection <SuggestedTransform> transforms,
ICollection <SuggestedTransform> transformsPostTrainer,
SuggestedTrainer trainer,
CacheBeforeTrainer cacheBeforeTrainerSettings)
{
var trainerInfo = trainer.BuildTrainer().Info;
AddNormalizationTransforms(context, trainerInfo, transforms);
var cacheBeforeTrainer = ShouldCacheBeforeTrainer(trainerInfo, cacheBeforeTrainerSettings);
return(new SuggestedPipeline(transforms, transformsPostTrainer, trainer, context, cacheBeforeTrainer));
}
/// <summary>
/// Given a predictor type, return a set of all permissible trainers (with their sweeper params, if defined).
/// </summary>
/// <returns>Array of viable learners.</returns>
public static IEnumerable <SuggestedTrainer> AllowedTrainers(MLContext mlContext, TaskKind task,
ColumnInformation columnInfo, IEnumerable <TrainerName> trainerWhitelist)
{
var trainerExtensions = TrainerExtensionCatalog.GetTrainers(task, trainerWhitelist, columnInfo);
var trainers = new List <SuggestedTrainer>();
foreach (var trainerExtension in trainerExtensions)
{
var learner = new SuggestedTrainer(mlContext, trainerExtension, columnInfo);
trainers.Add(learner);
}
return(trainers.ToArray());
}
public static SuggestedPipeline FromPipeline(MLContext context, Pipeline pipeline)
{
var transforms = new List <SuggestedTransform>();
var transformsPostTrainer = new List <SuggestedTransform>();
SuggestedTrainer trainer = null;
var trainerEncountered = false;
foreach (var pipelineNode in pipeline.Nodes)
{
if (pipelineNode.NodeType == PipelineNodeType.Trainer)
{
var trainerName = (TrainerName)Enum.Parse(typeof(TrainerName), pipelineNode.Name);
var trainerExtension = TrainerExtensionCatalog.GetTrainerExtension(trainerName);
var hyperParamSet = TrainerExtensionUtil.BuildParameterSet(trainerName, pipelineNode.Properties);
var columnInfo = TrainerExtensionUtil.BuildColumnInfo(pipelineNode.Properties);
trainer = new SuggestedTrainer(context, trainerExtension, columnInfo, hyperParamSet);
trainerEncountered = true;
}
else if (pipelineNode.NodeType == PipelineNodeType.Transform)
{
var estimatorName = (EstimatorName)Enum.Parse(typeof(EstimatorName), pipelineNode.Name);
var estimatorExtension = EstimatorExtensionCatalog.GetExtension(estimatorName);
var estimator = estimatorExtension.CreateInstance(context, pipelineNode);
var transform = new SuggestedTransform(pipelineNode, estimator);
if (!trainerEncountered)
{
transforms.Add(transform);
}
else
{
transformsPostTrainer.Add(transform);
}
}
}
return(new SuggestedPipeline(transforms, transformsPostTrainer, trainer, context, pipeline.CacheBeforeTrainer));
}
/// <summary>
/// Samples new hyperparameters for the trainer, and sets them.
/// Returns true if success (new hyperparameters were suggested and set). Else, returns false.
/// </summary>
private static bool SampleHyperparameters(MLContext context, SuggestedTrainer trainer,
IEnumerable <SuggestedPipelineRunDetail> history, bool isMaximizingMetric, IChannel logger)
{
try
{
var sps = ConvertToValueGenerators(trainer.SweepParams);
var sweeper = new SmacSweeper(context,
new SmacSweeper.Arguments
{
SweptParameters = sps
});
IEnumerable <SuggestedPipelineRunDetail> historyToUse = history
.Where(r => r.RunSucceeded && r.Pipeline.Trainer.TrainerName == trainer.TrainerName &&
r.Pipeline.Trainer.HyperParamSet != null &&
r.Pipeline.Trainer.HyperParamSet.Any() &&
FloatUtils.IsFinite(r.Score));
// get new set of hyperparameter values
var proposedParamSet = sweeper.ProposeSweeps(1, historyToUse.Select(h => h.ToRunResult(isMaximizingMetric))).FirstOrDefault();
if (!proposedParamSet.Any())
{
return(false);
}
// associate proposed parameter set with trainer, so that smart hyperparameter
// sweepers (like KDO) can map them back.
trainer.SetHyperparamValues(proposedParamSet);
return(true);
}
catch (Exception ex)
{
logger.Error($"SampleHyperparameters failed with exception: {ex}");
throw;
}
}
