public void FeatureNames()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var cols = dataSet.NumFeatures;
var names = new string[cols];
for (int i = 0; i < names.Length; ++i)
{
names[i] = string.Format("name{0}", i);
}
dataSet.SetFeatureNames(names);
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
using (var booster = new Booster(pms, dataSet))
{
Assert.Equal(1, booster.NumClasses);
Assert.Equal(names, booster.FeatureNames);
}
}
}
}
public void GetNumPredict_GetPredict()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
pms.Objective.Metric = MetricType.MultiLogLoss;
pms.Objective.NumClass = rand.Next(2, 4);
pms.Objective.Objective = ObjectiveType.MultiClass;
using (var booster = new Booster(pms, dataSet))
{
var numPredict = booster.GetNumPredict(0);
Assert.Equal(dataSet.NumRows * pms.Objective.NumClass, numPredict);
var rslts = booster.GetPredict(0);
Assert.Equal(rslts.Length, numPredict);
}
}
}
}
public void ResetTrainingData()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
using (var booster = new Booster(pms, dataSet))
{
// must have the same characteristics for binning to be able to do this.
var dataSet2 = new Dataset(dataSet, dataSet.NumRows / 2);
Assert.Equal(dataSet.NumFeatures, booster.NumFeatures);
Assert.Equal(dataSet.NumRows, booster.GetNumPredict(0));
booster.ResetTrainingData(dataSet2);
Assert.Equal(dataSet2.NumFeatures, booster.NumFeatures);
Assert.Equal(dataSet.NumRows / 2, dataSet2.NumRows);
Assert.Equal(dataSet2.NumRows, booster.GetNumPredict(0));
}
}
}
}
public void CheckDefaultParameters()
{
var rand = new Random();
using (var dataSet = DatasetTest.CreateRandom(rand, useDefaultDatasetParameters: true))
{
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
using (var booster = new Booster(pms, dataSet))
{
(var tree, var pmsout) = booster.GetModel();
if (pms != pmsout)
{
var dict = pms.ToDict();
var dictout = pmsout.ToDict();
var keys = dict.Keys.Concat(dictout.Keys).Distinct();
foreach (var key in keys)
{
dict.TryGetValue(key, out string valIn);
dictout.TryGetValue(key, out string valOut);
if (valIn != valOut)
{
throw (new Exception($"Default value of {key} mismatch: {valIn} vs {valOut}"));
}
}
}
}
}
}
public static DataSparse Dense2Sparse(DataDense data)
{
return((data == null) ? null : new DataSparse()
{
Features = DatasetTest.Dense2Sparse(data.Features),
Labels = data.Labels,
Weights = data.Weights,
Groups = data.Groups
});
}
public void GetEvalCounts_GetEvalNames()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
var metric = CreateRandomMetric(rand);
if (metric != MetricType.Ndcg && metric != MetricType.Map && (pms.Objective.NumClass > 1 || (metric != MetricType.MultiLogLoss && metric != MetricType.MultiError && metric != MetricType.AucMu)))
{
if (metric == MetricType.Gamma ||
metric == MetricType.GammaDeviance ||
metric == MetricType.Tweedie ||
metric == MetricType.Poisson)
{
var labels = dataSet.GetLabels();
for (var i = 0; i < labels.Length; i++)
{
labels[i] = (float)Math.Max(1e-3, Math.Abs(labels[i]));
}
dataSet.SetLabels(labels);
}
pms.Objective.Metric = metric;
//pms.Metric.IsProvideTrainingMetric = true;
using (var booster = new Booster(pms, dataSet))
{
var numEval = booster.EvalCounts;
var evalNames = booster.EvalNames;
Assert.Equal(numEval, evalNames.Length);
if (numEval > 0)
{
Assert.Equal(1, numEval);
if (metric == MetricType.DefaultMetric)
{
var metric2 = EnumHelper.ParseMetric(EnumHelper.GetObjectiveString(pms.Objective.Objective));
Assert.Equal(metric2, evalNames[0]);
}
else
{
Assert.Equal(metric, evalNames[0]);
}
}
}
}
}
}
}
public void Create()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters();
using (var booster = new Booster(pms, dataSet))
{
}
}
}
}
public void NumFeatures()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters();
using (var booster = new Booster(pms, dataSet))
{
Assert.Equal(booster.NumFeatures, dataSet.NumFeatures);
}
}
}
}
public void NumClasses()
{
var rand = new Random();
for (int test = 0; test < 100; ++test)
{
using (var dataSet = DatasetTest.CreateRandom(rand))
{
var pms = new Parameters();
pms.Objective.Metric = MetricType.MultiLogLoss;
pms.Objective.NumClass = rand.Next(2, 4);
pms.Objective.Objective = ObjectiveType.MultiClass;
using (var booster = new Booster(pms, dataSet))
{
Assert.Equal(pms.Objective.NumClass, booster.NumClasses);
}
}
}
}
public void CheckDefaultParameters()
{
var rand = new Random();
using (var dataSet = DatasetTest.CreateRandom(rand, useDefaultDatasetParameters: true))
{
var pms = new Parameters()
{
Common = dataSet.CommonParameters, Dataset = dataSet.DatasetParameters
};
using (var booster = new Booster(pms, dataSet))
{
(var tree, var pmsout) = booster.GetModel();
if (pms != pmsout)
{
// if we use DefaultMetric, the returned parameter set will return the objective as the metric
if (pms.Objective.Metric == MetricType.DefaultMetric)
{
pms.Objective.Metric = EnumHelper.ParseMetric(EnumHelper.GetObjectiveString(pms.Objective.Objective));
}
var dict = pms.ToDict();
var dictout = pmsout.ToDict();
var keys = dict.Keys.Concat(dictout.Keys).Distinct();
foreach (var key in keys)
{
dict.TryGetValue(key, out string valIn);
dictout.TryGetValue(key, out string valOut);
if (valIn != valOut)
{
throw (new Exception($"Default value of {key} mismatch: {valIn} vs {valOut}"));
}
}
}
}
}
}