internal static void DisposeIfDisposable <TPrediction>(IPredictorModel <TPrediction> model)
{
if (model is IDisposable)
{
((IDisposable)model).Dispose();
}
}
/// <summary>
/// Learns a regression ensemble on the provided indices
/// </summary>
/// <param name="observations"></param>
/// <param name="targets"></param>
/// <param name="indices"></param>
/// <returns></returns>
public RegressionEnsembleModel Learn(F64Matrix observations, double[] targets,
int[] indices)
{
Checks.VerifyObservationsAndTargets(observations, targets);
Checks.VerifyIndices(indices, observations, targets);
var ensembleModels = new IPredictorModel <double> [m_learners.Length];
var sampleSize = (int)Math.Round(m_subSampleRatio * indices.Length);
if (sampleSize < 1)
{
throw new ArgumentException("subSampleRatio two small");
}
var inSample = new int[sampleSize];
for (int i = 0; i < m_learners.Length; i++)
{
Trace.WriteLine("Training model: " + (i + 1));
if (m_subSampleRatio != 1.0)
{
Sample(inSample, indices);
ensembleModels[i] = m_learners[i](observations, targets, inSample);
}
else
{
ensembleModels[i] = m_learners[i](observations, targets, indices);
}
}
return(new RegressionEnsembleModel(ensembleModels, m_ensembleStrategy()));
}
public override List <IDictionary <string, string> > ReleaseSets(IPredictorModel model)
{
return(new List <IDictionary <string, string> >()
{
model.SelectedReleases
});
}
public PredictorPresenter(IPredictorModel model, IPredictorView view)
{
this.model = model;
this.view = view;
model.OnTitleUpdated += x => view.Title = x;
model.OnClearReport += () => view.ClearReport();
model.OnAddReport += x => view.AddReport(x);
model.OnClearRocList += () => view.ClearROCs();
model.OnRocAdded += view.AddRoc;
model.OnReadyStateChanged += x => view.Ready = x;
model.OnProgressStateChanged += x => view.Status = x;
model.OnError += x => view.ShowError(x);
view.OnOpenConfigFile += OpenConfigFile;
view.OnPredict += Predict;
view.OnShowROC += model.ShowROC;
view.ShowFiles = model.ShowFiles;
view.Evaluate = model.Evaluate;
view.EvaluateUsingROC = model.EvaluateUsingROC;
view.EvaluateRanking = model.EvaluateRanking;
view.ShowEstimationStats = model.ShowEstimationStats;
view.ShowTotalResult = model.ShowTotalResult;
if (model.ReleaseSetGetting == ReleaseSetGettingAlgo.All)
{
view.ReleaseSetGettingAll = true;
}
else if (model.ReleaseSetGetting == ReleaseSetGettingAlgo.IncrementalGrowth)
{
view.ReleaseSetGettingIncrementalGrowth = true;
}
else
{
view.ReleaseSetGettingFixed = true;
}
view.ReleaseSetSize = model.ReleaseSetSize;
}
public PredictorPresenter(IPredictorModel model, IPredictorView view)
{
this.model = model;
this.view = view;
model.OnTitleUpdated += x => view.Title = x;
model.OnClearReport += () => view.ClearReport();
model.OnAddReport += x => view.AddReport(x);
model.OnClearRocList += () => view.ClearROCs();
model.OnRocAdded += view.AddRoc;
model.OnReadyStateChanged += x => view.Ready = x;
model.OnProgressStateChanged += x => view.Status = x;
model.OnError += x => view.ShowError(x);
view.OnOpenConfigFile += OpenConfigFile;
view.OnPredict += Predict;
view.OnShowROC += model.ShowROC;
view.ShowFiles = model.ShowFiles;
view.Evaluate = model.Evaluate;
view.EvaluateUsingROC = model.EvaluateUsingROC;
view.EvaluateRanking = model.EvaluateRanking;
view.ShowEstimationStats = model.ShowEstimationStats;
view.ShowTotalResult = model.ShowTotalResult;
if (model.ReleaseSetGetting == ReleaseSetGettingAlgo.All)
{
view.ReleaseSetGettingAll = true;
}
else if (model.ReleaseSetGetting == ReleaseSetGettingAlgo.IncrementalGrowth)
{
view.ReleaseSetGettingIncrementalGrowth = true;
}
else
{
view.ReleaseSetGettingFixed = true;
}
view.ReleaseSetSize = model.ReleaseSetSize;
}
/// <summary>
/// Saves the PredictorModel to the given path
/// </summary>
private void SavePredictorModel(IPredictorModel pm, string path)
{
Contracts.CheckValue(pm, nameof(pm));
using (var handle = _host.CreateOutputFile(path))
using (var fs = handle.CreateWriteStream())
pm.Save(_host, fs);
}
internal static void DisposeIfDisposable <TPrediction>(IPredictorModel <TPrediction> model)
{
var modelDisposable = typeof(IDisposable).IsAssignableFrom(model.GetType());
if (modelDisposable)
{
((IDisposable)model).Dispose();
}
}
/// <summary>
///
/// </summary>
/// <param name="ensembleModels">Models included in the ensemble</param>
/// <param name="metaModel">Meta or top level model to combine the ensemble models</param>
/// <param name="includeOriginalFeaturesForMetaLearner">True; the meta learner also receives the original features.
/// False; the meta learner only receives the output of the ensemble models as features</param>
/// <param name="numberOfClasses">Number of classes in the classification problem</param>
public ClassificationStackingEnsembleModel(IPredictorModel <ProbabilityPrediction>[] ensembleModels,
IPredictorModel <ProbabilityPrediction> metaModel,
bool includeOriginalFeaturesForMetaLearner,
int numberOfClasses)
{
m_ensembleModels = ensembleModels ?? throw new ArgumentException(nameof(ensembleModels));
m_metaModel = metaModel ?? throw new ArgumentException(nameof(metaModel));
m_includeOriginalFeaturesForMetaLearner = includeOriginalFeaturesForMetaLearner;
m_numberOfClasses = numberOfClasses;
}
public override List <IDictionary <string, string> > ReleaseSets(IPredictorModel model)
{
List <IDictionary <string, string> > list = new List <IDictionary <string, string> >();
for (int i = 1; i <= model.SelectedReleases.Count; i++)
{
list.Add(
model.SelectedReleases.Take(i).ToDictionary(x => x.Key, x => x.Value)
);
}
return(list);
}
/// <summary>
///
/// </summary>
/// <param name="ensembleModels">Models included in the ensemble</param>
/// <param name="metaModel">Meta or top level model to combine the ensemble models</param>
/// <param name="includeOriginalFeaturesForMetaLearner">True; the meta learner also recieves the original features.
/// False; the meta learner only recieves the output of the ensemble models as features</param>
public RegressionStackingEnsembleModel(IPredictorModel <double>[] ensembleModels, IPredictorModel <double> metaModel, bool includeOriginalFeaturesForMetaLearner)
{
if (ensembleModels == null)
{
throw new ArgumentException("ensembleModels");
}
if (metaModel == null)
{
throw new ArgumentException("metaModel");
}
m_ensembleModels = ensembleModels;
m_metaModel = metaModel;
m_includeOriginalFeaturesForMetaLearner = includeOriginalFeaturesForMetaLearner;
}
public void LoadModel(string path)
{
model = null;
try
{
var binarySerializer = new GenericBinarySerializer();
model = binarySerializer.Deserialize <IPredictorModel <double> >(
() => new StreamReader(root + path));
}
catch (Exception e)
{
Debug.Log(e.Message);
model = null;
}
}
/// <summary>
///
/// </summary>
/// <param name="ensembleModels">Models included in the ensemble</param>
/// <param name="metaModel">Meta or top level model to combine the ensemble models</param>
/// <param name="includeOriginalFeaturesForMetaLearner">True; the meta learner also recieves the original features.
/// False; the meta learner only recieves the output of the ensemble models as features</param>
/// <param name="numberOfClasses">Number of classes in the classification problem</param>
public ClassificationStackingEnsembleModel(IPredictorModel <ProbabilityPrediction>[] ensembleModels, IPredictorModel <ProbabilityPrediction> metaModel,
bool includeOriginalFeaturesForMetaLearner, int numberOfClasses)
{
if (ensembleModels == null)
{
throw new ArgumentException("ensembleModels");
}
if (metaModel == null)
{
throw new ArgumentException("metaModel");
}
m_ensembleModels = ensembleModels;
m_metaModel = metaModel;
m_includeOriginalFeaturesForMetaLearner = includeOriginalFeaturesForMetaLearner;
m_numberOfClasses = numberOfClasses;
}
public void TrainModelWithRangeOfPossibleParametersTest()
{
const string baseFolder = nameof(TrainModelWithRangeOfPossibleParametersTest);
Directory.CreateDirectory(baseFolder);
var generator = new ModelTrainingFilesGenerator();
// create catalog items
IList <SarCatalogItem> catalogItems;
string[] featureNames;
string catalogFilePath = Path.Combine(baseFolder, "catalog.csv");
generator.CreateCatalogFile(catalogFilePath);
var itemIdsIndex = new ConcurrentDictionary <string, uint>();
var catalogParser = new CatalogFileParser(0, itemIdsIndex);
var parsingReport = catalogParser.ParseCatalogFile(catalogFilePath, CancellationToken.None, out catalogItems, out featureNames);
Assert.IsTrue(parsingReport.IsCompletedSuccessfuly);
// create usage items
IList <SarUsageEvent> usageEvents;
string usageFileFolderPath = Path.Combine(baseFolder, "usage");
Directory.CreateDirectory(usageFileFolderPath);
generator.CreateUsageFile(Path.Combine(usageFileFolderPath, "usage.csv"), 10000);
var userIdsIndex = new ConcurrentDictionary <string, uint>();
var usageFilesParser = new UsageEventsFilesParser(itemIdsIndex, userIdsIndex);
parsingReport = usageFilesParser.ParseUsageEventFiles(usageFileFolderPath, CancellationToken.None, out usageEvents);
Assert.IsTrue(parsingReport.IsCompletedSuccessfuly);
int count = 0;
var sarTrainer = new SarTrainer();
IDictionary <string, double> catalogFeatureWeights;
foreach (IModelTrainerSettings settings in GetAllModelTrainingParameters())
{
IPredictorModel model = sarTrainer.Train(settings, usageEvents, catalogItems, featureNames, userIdsIndex.Count, itemIdsIndex.Count, out catalogFeatureWeights);
Assert.IsNotNull(model, $"Expected training to complete successfully when using settings#{count}: {settings}");
count++;
}
}
private void CalibrateSpectra(IPredictorModel <double> ms1predictor, IPredictorModel <double> ms2predictor)
{
Parallel.ForEach(Partitioner.Create(1, MyMsDataFile.NumSpectra + 1), new ParallelOptions {
MaxDegreeOfParallelism = commonParameters.MaxThreadsToUsePerFile
}, (fff, loopState) =>
{
for (int i = fff.Item1; i < fff.Item2; i++)
{
// Stop loop if canceled
if (GlobalVariables.StopLoops)
{
loopState.Stop();
return;
}
var scan = MyMsDataFile.GetOneBasedScan(i);
if (scan.MsnOrder == 2)
{
var precursorScan = MyMsDataFile.GetOneBasedScan(scan.OneBasedPrecursorScanNumber.Value);
if (!scan.SelectedIonMonoisotopicGuessIntensity.HasValue && scan.SelectedIonMonoisotopicGuessMz.HasValue)
{
scan.ComputeMonoisotopicPeakIntensity(precursorScan.MassSpectrum);
}
double theFunc(MzPeak x) => x.Mz - ms2predictor.Predict(new double[] { x.Mz, scan.RetentionTime, Math.Log(scan.TotalIonCurrent), scan.InjectionTime.HasValue ? Math.Log(scan.InjectionTime.Value) : double.NaN, Math.Log(x.Intensity) });
double theFuncForPrecursor(MzPeak x) => x.Mz - ms1predictor.Predict(new double[] { x.Mz, precursorScan.RetentionTime, Math.Log(precursorScan.TotalIonCurrent), precursorScan.InjectionTime.HasValue ? Math.Log(precursorScan.InjectionTime.Value) : double.NaN, Math.Log(x.Intensity) });
scan.TransformMzs(theFunc, theFuncForPrecursor);
}
else
{
Func <MzPeak, double> theFunc = x => x.Mz - ms1predictor.Predict(new double[] { x.Mz, scan.RetentionTime, Math.Log(scan.TotalIonCurrent), scan.InjectionTime.HasValue ? Math.Log(scan.InjectionTime.Value) : double.NaN, Math.Log(x.Intensity) });
scan.MassSpectrum.ReplaceXbyApplyingFunction(theFunc);
}
}
}
);
}
public override List <IDictionary <string, string> > ReleaseSets(IPredictorModel model)
{
List <IDictionary <string, string> > list = new List <IDictionary <string, string> >();
if (model.SelectedReleases.Count <= model.ReleaseSetSize)
{
list.Add(model.SelectedReleases);
}
else
{
for (int i = 0; i <= model.SelectedReleases.Count - model.ReleaseSetSize; i++)
{
list.Add(
model.SelectedReleases.Skip(i).Take(model.ReleaseSetSize)
.ToDictionary(x => x.Key, x => x.Value)
);
}
}
return(list);
}
/// <summary>
/// Creates a new instance of the <see cref="TrainedModel"/> class.
/// </summary>
/// <param name="predictorModel">The predictor model</param>
/// <param name="properties">The model properties</param>
/// <param name="itemIdIndex">The indexed item ids </param>
internal TrainedModel(IPredictorModel predictorModel, ModelProperties properties, string[] itemIdIndex)
{
if (predictorModel == null)
{
throw new ArgumentNullException(nameof(predictorModel));
}
if (properties == null)
{
throw new ArgumentNullException(nameof(properties));
}
if (itemIdIndex == null)
{
throw new ArgumentNullException(nameof(itemIdIndex));
}
_predictorModel = predictorModel;
RecommenderData = new ModelRecommenderData(_predictorModel.Predictor as IUserHistoryToItemsRecommender);
Properties = properties;
ItemIdIndex = itemIdIndex;
}
public abstract List <IDictionary <string, string> > ReleaseSets(IPredictorModel model);
/// <summary>
/// Trains a model using the input files
/// </summary>
/// <param name="settings">The trainer settings</param>
/// <param name="workFolderPath">A temp work folder for storing intermediate files</param>
/// <param name="usageFolderPath">The path to the folder of usage files</param>
/// <param name="catalogFilePath">The path to the catalog file</param>
/// <param name="evaluationFolderPath">The path to the evaluation file (optional) </param>
/// <param name="cancellationToken">A cancellation token used to abort the training</param>
private ModelTrainResult TrainModelInternal(IModelTrainerSettings settings, string workFolderPath, string usageFolderPath,
string catalogFilePath, string evaluationFolderPath, CancellationToken cancellationToken)
{
var duration = ModelTraininigDuration.Start();
var result = new ModelTrainResult {
Duration = duration
};
var userIdsIndexMap = new ConcurrentDictionary <string, uint>();
var itemIdsIndexMap = new ConcurrentDictionary <string, uint>();
// parse the catalog file
IList <SarCatalogItem> catalogItems = null;
string[] catalogFeatureNames = null;
if (!string.IsNullOrWhiteSpace(catalogFilePath) && File.Exists(catalogFilePath))
{
// report progress
_progressMessageReportDelegate("Parsing Catalog File");
// create a catalog file parser
var catalogParser = new CatalogFileParser(MaximumParsingErrorsCount, itemIdsIndexMap, _tracer);
// parse the catalog file
result.CatalogFilesParsingReport = catalogParser.ParseCatalogFile(catalogFilePath, cancellationToken,
out catalogItems, out catalogFeatureNames);
// record the catalog parsing duration
duration.SetCatalogParsingDuration();
_tracer.TraceInformation($"Catalog parsing completed in {duration.CatalogParsingDuration.TotalMinutes} minutes");
// get the catalog items count
result.CatalogItemsCount = catalogItems.Count;
// fail the training if parsing had failed or yielded no items
if (!result.CatalogFilesParsingReport.IsCompletedSuccessfuly || !catalogItems.Any())
{
result.CompletionMessage = "Failed to parse catalog file or parsing found no valid items";
_tracer.TraceInformation(result.CompletionMessage);
return(result);
}
// clear the catalog items list if it's not used anymore
if (!settings.EnableColdItemPlacement)
{
catalogItems.Clear();
}
}
// report progress
_progressMessageReportDelegate("Parsing Usage Events Files");
// create a usage events files parser that skips events of unknown item ids (if catalog was provided))
var usageEventsParser = new UsageEventsFilesParser(itemIdsIndexMap, userIdsIndexMap,
MaximumParsingErrorsCount, catalogItems != null, _tracer);
_tracer.TraceInformation("Parsing the usage event files");
IList <SarUsageEvent> usageEvents;
result.UsageFilesParsingReport =
usageEventsParser.ParseUsageEventFiles(usageFolderPath, cancellationToken, out usageEvents);
// record the usage files parsing duration
duration.SetUsageFilesParsingDuration();
_tracer.TraceInformation($"Usage file(s) parsing completed in {duration.UsageFilesParsingDuration.TotalMinutes} minutes");
// fail the training if parsing had failed or yielded no events
if (!result.UsageFilesParsingReport.IsCompletedSuccessfuly || !usageEvents.Any())
{
result.CompletionMessage = "Failed to parse usage file(s) or parsing found no valid items";
_tracer.TraceInformation(result.CompletionMessage);
return(result);
}
_tracer.TraceInformation($"Found {userIdsIndexMap.Count} unique users");
result.UniqueUsersCount = userIdsIndexMap.Count;
_tracer.TraceInformation($"Found {itemIdsIndexMap.Count} unique items");
result.UniqueItemsCount = usageEvents.Select(x => x.ItemId).Distinct().Count();
_tracer.TraceInformation("Extracting the indexed item ids from the item index map");
string[] itemIdsIndex = itemIdsIndexMap.OrderBy(kvp => kvp.Value).Select(kvp => kvp.Key).ToArray();
_tracer.TraceInformation($"Sorting the usage events based on the cooccurrenceUnit unit ({settings.CooccurrenceUnit})");
switch (settings.CooccurrenceUnit)
{
case CooccurrenceUnit.User:
usageEvents = usageEvents.OrderBy(x => x.UserId).ToArray();
break;
case CooccurrenceUnit.Timestamp:
usageEvents = usageEvents.OrderBy(x => x.Timestamp).ThenBy(x => x.UserId).ToArray();
break;
}
_tracer.TraceInformation("Finished sorting usage events.");
Stopwatch storeUserHistoryDuration = null;
Task storeUserHistoryTask = null;
if (settings.EnableUserToItemRecommendations && _userHistoryStore != null)
{
storeUserHistoryDuration = Stopwatch.StartNew();
_tracer.TraceInformation($"Extracting the indexed user ids from the user index map ({userIdsIndexMap.Count:N} users)");
string[] userIdsIndex = userIdsIndexMap.OrderBy(kvp => kvp.Value).Select(kvp => kvp.Key).ToArray();
_tracer.TraceInformation($"Asynchronously starting to store usage events per user (total of {usageEvents.Count:N} items)");
storeUserHistoryTask = Task.Run(() =>
_userHistoryStore.StoreUserHistoryEventsAsync(usageEvents, userIdsIndex, cancellationToken), cancellationToken);
}
// if provided, parse the evaluation usage event files
int evaluationUsageEventsCount = 0;
string parsedEvaluationUsageEventsFilePath = null;
if (!string.IsNullOrWhiteSpace(evaluationFolderPath) && Directory.Exists(evaluationFolderPath))
{
// report progress
_progressMessageReportDelegate("Parsing Evaluation Usage Events Files");
_tracer.TraceInformation("Parsing the evaluation usage event files");
IList <SarUsageEvent> evaluationUsageEvents;
result.EvaluationFilesParsingReport = usageEventsParser.ParseUsageEventFiles(evaluationFolderPath,
cancellationToken, out evaluationUsageEvents);
if (result.EvaluationFilesParsingReport.IsCompletedSuccessfuly)
{
// set the evaluation usage events count
evaluationUsageEventsCount = evaluationUsageEvents.Count;
_tracer.TraceInformation("Storing the parsed usage events for evaluation to reduce memory print");
parsedEvaluationUsageEventsFilePath = Path.Combine(workFolderPath, Path.GetTempFileName());
File.WriteAllLines(parsedEvaluationUsageEventsFilePath,
evaluationUsageEvents.Select(JsonConvert.SerializeObject));
}
else
{
_tracer.TraceWarning("Skipping model evaluation as it failed to parse evaluation usage files.");
}
// record the evaluation usage files parsing duration
duration.SetEvaluationUsageFilesParsingDuration();
_tracer.TraceInformation($"Evaluation usage file(s) parsing completed in {duration.EvaluationUsageFilesParsingDuration.TotalMinutes} minutes");
}
// clear the indices maps as they are no longer needed
userIdsIndexMap.Clear();
itemIdsIndexMap.Clear();
cancellationToken.ThrowIfCancellationRequested();
// report progress
_progressMessageReportDelegate("Core Training");
_tracer.TraceInformation("Training a new model using SAR trainer");
IDictionary <string, double> catalogFeatureWeights;
var sarTrainer = new SarTrainer(_tracer);
IPredictorModel sarModel = sarTrainer.Train(settings, usageEvents, catalogItems, catalogFeatureNames, result.UniqueUsersCount,
result.CatalogItemsCount ?? result.UniqueItemsCount, out catalogFeatureWeights, cancellationToken);
_tracer.TraceInformation("SAR training was completed.");
// create the trained model properties
var modelProperties = new ModelProperties
{
IncludeHistory = settings.AllowSeedItemsInRecommendations,
EnableUserAffinity = settings.EnableUserAffinity,
IsUserToItemRecommendationsSupported = settings.EnableUserToItemRecommendations,
Decay = TimeSpan.FromDays(settings.DecayPeriodInDays),
ReferenceDate = usageEventsParser.MostRecentEventTimestamp,
UniqueUsersCount = result.UniqueUsersCount,
};
// create the trained model
result.Model = new TrainedModel(sarModel, modelProperties, itemIdsIndex);
// set the catalog features weights
result.CatalogFeatureWeights = catalogFeatureWeights;
// record the core training duration
duration.SetTrainingDuration();
// run model evaluation if evaluation usage event are available
if (evaluationUsageEventsCount > 0 && parsedEvaluationUsageEventsFilePath != null)
{
// report progress
_progressMessageReportDelegate("Evaluating Trained Model");
var evaluationUsageEvents = new List <SarUsageEvent>(evaluationUsageEventsCount);
// load the evaluation usage events
using (var reader = new StreamReader(parsedEvaluationUsageEventsFilePath))
{
while (!reader.EndOfStream)
{
evaluationUsageEvents.Add(JsonConvert.DeserializeObject <SarUsageEvent>(reader.ReadLine()));
}
}
_tracer.TraceInformation("Starting model evaluation");
var evaluator = new ModelEvaluator(_tracer);
result.ModelMetrics = evaluator.Evaluate(result.Model, usageEvents, evaluationUsageEvents, cancellationToken);
// record the evaluation duration
duration.SetEvaluationDuration();
}
if (storeUserHistoryTask != null)
{
_tracer.TraceInformation("Waiting for storing of usage events per user (user history) to complete");
if (!storeUserHistoryTask.IsCompleted)
{
_progressMessageReportDelegate("Storing User History");
// set the reporting flag to true so usage history upload progress will get reported to model status
_reportUserHistoryProgress = true;
}
try
{
storeUserHistoryTask.Wait(cancellationToken);
storeUserHistoryDuration?.Stop();
duration.StoringUserHistoryDuration = storeUserHistoryDuration?.Elapsed;
_tracer.TraceInformation(
$"Storing usage events per user (user history) to complete after {duration.StoringUserHistoryDuration.Value.TotalMinutes} minutes");
}
catch (AggregateException ex)
{
var exception = new Exception("Exception while trying to store user history", ex);
_tracer.TraceError(exception.ToString());
throw exception;
}
}
// stop measuring the duration and record the total duration
duration.Stop();
// return the train result
result.CompletionMessage = "Model Training Completed Successfully";
return(result);
}
private static void SavePredictorModelToFile(IPredictorModel model, string path, IHost host)
{
using (var fs = File.OpenWrite(path))
model.Save(host, fs);
}
/// <summary>
/// Trains a model using SAR.
/// </summary>
/// <param name="settings">The training settings</param>
/// <param name="usageEvents">The usage events to use for training</param>
/// <param name="catalogItems">The catalog items to use for training</param>
/// <param name="featureNames">The names of the catalog items features, in the same order as the feature values in the catalog</param>
/// <param name="uniqueUsersCount">The number of users in the user id index file.</param>
/// <param name="uniqueUsageItemsCount">The number of usage items in the item id index file</param>
/// <param name="catalogFeatureWeights">The computed catalog items features weights (if relevant)</param>
/// <param name="cancellationToken">A cancellation token</param>
public IPredictorModel Train(ITrainingSettings settings,
IList <SarUsageEvent> usageEvents,
IList <SarCatalogItem> catalogItems,
string[] featureNames,
int uniqueUsersCount,
int uniqueUsageItemsCount,
out IDictionary <string, double> catalogFeatureWeights,
CancellationToken cancellationToken)
{
if (settings == null)
{
throw new ArgumentNullException(nameof(settings));
}
if (usageEvents == null)
{
throw new ArgumentNullException(nameof(usageEvents));
}
if (settings.EnableColdItemPlacement && catalogItems == null)
{
throw new ArgumentNullException(nameof(catalogItems));
}
if (uniqueUsersCount < 0)
{
var exception = new ArgumentException($"{nameof(uniqueUsersCount)} must be a positive integer");
_tracer.TraceWarning(exception.ToString());
throw exception;
}
if (uniqueUsageItemsCount < 0)
{
var exception = new ArgumentException($"{nameof(uniqueUsageItemsCount)} must be a positive integer");
_tracer.TraceWarning(exception.ToString());
throw exception;
}
cancellationToken.ThrowIfCancellationRequested();
using (TlcEnvironment environment = new TlcEnvironment(verbose: true))
{
_detectedFeatureWeights = null;
try
{
environment.AddListener <ChannelMessage>(ChannelMessageListener);
IHost environmentHost = environment.Register("SarHost");
// bind the cancellation token to SAR cancellation
using (cancellationToken.Register(() => { environmentHost.StopExecution(); }))
{
_tracer.TraceInformation("Starting training model using SAR");
IPredictorModel model = TrainModel(environmentHost, settings, usageEvents, catalogItems, uniqueUsersCount,
uniqueUsageItemsCount);
catalogFeatureWeights = new Dictionary <string, double>();
if (_detectedFeatureWeights != null && featureNames != null)
{
if (_detectedFeatureWeights.Length == featureNames.Length)
{
for (int i = 0; i < featureNames.Length; i++)
{
catalogFeatureWeights[featureNames[i]] = _detectedFeatureWeights[i];
}
}
else
{
_tracer.TraceWarning(
$"Found a mismatch between number of feature names ({featureNames.Length}) and the number of feature weights ({_detectedFeatureWeights.Length})");
}
}
return(model);
}
}
finally
{
environment.RemoveListener <ChannelMessage>(ChannelMessageListener);
}
}
}
/// <summary>
///
/// </summary>
/// <param name="ensembleModels">Models included in the ensemble</param>
/// <param name="metaModel">Meta or top level model to combine the ensemble models</param>
/// <param name="includeOriginalFeaturesForMetaLearner">True; the meta learner also receives the original features.
/// False; the meta learner only receives the output of the ensemble models as features</param>
public RegressionStackingEnsembleModel(IPredictorModel <double>[] ensembleModels, IPredictorModel <double> metaModel, bool includeOriginalFeaturesForMetaLearner)
{
m_ensembleModels = ensembleModels ?? throw new ArgumentException(nameof(ensembleModels));
m_metaModel = metaModel ?? throw new ArgumentException(nameof(metaModel));
m_includeOriginalFeaturesForMetaLearner = includeOriginalFeaturesForMetaLearner;
}
public void Train(string dataName, string savePath)
{
var parser = new CsvParser(() => new StreamReader(root + dataName + ".csv"), ',');
var targetName = "f5";
var enemyName = "f6";
try
{
// read data
var observations = parser.EnumerateRows(c => c != targetName && c != enemyName)
.ToF64Matrix();
var targets = parser.EnumerateRows(targetName)
.ToF64Vector();
//Debug.Log("READ");
// split data
//var splitter = new StratifiedTrainingTestIndexSplitter<double>(trainingPercentage: 0.9, seed: 1);
//var trainingTestSplit = splitter.SplitSet(observations, targets);
//var trainingSet = trainingTestSplit.TrainingSet;
//var testSet = trainingTestSplit.TestSet;
//Debug.Log("SPLIT");
//model = learner.Learn(trainingSet.Observations, trainingSet.Targets);
model = learner.Learn(observations, targets);
//Debug.Log("TRAINED");
//var met1 = new MeanSquaredErrorRegressionMetric();
//var met2 = new CoefficientOfDeterminationMetric();
//var met3 = new MeanAbsolutErrorRegressionMetric();
//var met4 = new NormalizedGiniCoefficientRegressionMetric();
//var met5 = new RootMeanSquarePercentageRegressionMetric();
//var pred = model.Predict(testSet.Observations);
//Debug.Log(met1.Error(testSet.Targets, pred) + " <-> " + type);
//Debug.Log(met2.Error(testSet.Targets, pred) + " <-> " + type);
//Debug.Log(met3.Error(testSet.Targets, pred) + " <-> " + type);
//Debug.Log(met4.Error(testSet.Targets, pred) + " <-> " + type);
//Debug.Log(met5.Error(testSet.Targets, pred) + " <-> " + type);
// NEURAL TEST
Debug.Log("TRAINED");
// save
SaveModel(savePath);
// load
//LoadModel("" + LevelData.instance.stage + "/" + LevelData.instance.playerID + ".model");
}
catch (Exception e)
{
Debug.Log(e.Message);
model = null;
}
finally
{
}
}
