public void testing_saving_and_loading_saved_model()
{
// Save
Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv")).
Classifiers.Trees.RandomForest.
NumExecutionSlots(4).
NumFeatures(5).
NumTrees(50).
FlushToFile("titanic_randor_forest.model");
// Load
IUntypedBaseClassifier <weka.classifiers.Classifier> classifier = BaseClassifier.Read("titanic_randor_forest.model");
TitanicDataRow row = new TitanicDataRow
{
age = 10,
pclass = "1",
sex = "male",
embarked = "C"
};
// Classify
double prediction = classifier.ClassifyInstance(Runtime.BuildInstance(0, row));
double proba = classifier.ClassifyInstanceProba(Runtime.BuildInstance(0, row));
Assert.AreEqual(0.0, prediction);
Assert.IsTrue(proba < 0.5);
}
public void test_basic_unsupervised_attribute_filtering()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
Runtime newrt = rt.Filters.UnsupervisedAttribute.RemoveByName.
Expression("sex").
RunFilter();
Assert.AreEqual(rt.NumAttributes, newrt.NumAttributes + 1);
}
public void test_basic_unsupervised_instance_filtering()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
Runtime newrt = rt.Filters.UnsupervisedInstance.RemoveRange.
InstancesIndices("1-100").
RunFilter();
Assert.AreEqual(rt.NumInstances, newrt.NumInstances + 100);
}
public void test_basic_supervised_attribute_filtering()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
Runtime newrt = rt.Filters.SupervisedAttribute.NominalToBinary.
AttributeIndices("2"). // pcclass (1,2,3)
RunFilter();
// Add 3 new binary attributes and remove orignial
Assert.AreEqual(rt.NumAttributes + 3 - 1, newrt.NumAttributes);
}
public void test_simple_evaluation()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
rt.Classifiers.Trees.RandomForest.
NumExecutionSlots(4).
NumFeatures(5).
NumTrees(50).
EvaluateWithCrossValidation();
}
public void test_basic_supervised_instance_filtering()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
Runtime newrt = rt.Filters.SupervisedInstance.SpreadSubsample.
DistributionSpread(1.0).
RunFilter();
Assert.AreEqual(rt.NumInstances, 891);
Assert.AreEqual(newrt.NumInstances, 684);
Assert.AreEqual(newrt.Count(i => i.ClassValue == 0.0), 342);
Assert.AreEqual(newrt.Count(i => i.ClassValue == 1.0), 342);
}
public void simple_attribute_selection_tests_with_indexes()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
BestFirst alg = rt.AttributeSelections.Algorithms.BestFirst.
Direction(BestFirst.EDirection.Bi_directional).
LookupCacheSize(10);
PicNetML.AttrSel.Evals.CfsSubset eval = rt.AttributeSelections.Evaluators.CfsSubset.
LocallyPredictive(true).
MissingSeparate(true);
int[] indexes = alg.SearchIndexes(eval);
Assert.AreEqual(new[] { 2, 0 }, indexes);
}
public void test_building_predictions_lines()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
RandomForest cls = rt.Classifiers.Trees.RandomForest.
NumExecutionSlots(4).
NumFeatures(5).
NumTrees(50);
Runtime testset = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_test.csv"), preprocessor: TestLinePreproc);
int count = testset.NumInstances;
System.Collections.Generic.List <string> lines = testset.GeneratePredictions(GeneratePredictionLine, cls);
Assert.AreEqual(count, lines.Count);
}
[Test] public void simple_attribute_selection_tests_with_new_runtime()
{
var rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
var alg = rt.AttributeSelections.Algorithms.BestFirst.
Direction(BestFirst.EDirection.Bi_directional).
LookupCacheSize(10);
var eval = rt.AttributeSelections.Evaluators.CfsSubset.
LocallyPredictive(true).
MissingSeparate(true);
var newrt = alg.Search(eval);
var names = newrt.EnumerateAttributes.Select(a => a.Name).ToArray();
Assert.AreEqual(new[] { "sex", "survived" }, names);
}
public void test_basic_clustering()
{
Runtime rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
// Remove the classifier (which upsets clusterers)
rt = rt.Filters.UnsupervisedAttribute.Remove.AttributeIndices("1").RunFilter();
PicNetML.Clstr.IBaseClusterer <weka.clusterers.SimpleKMeans> clusterer = rt.Clusterers.SimpleKMeans.
NumClusters(10).
Build();
Assert.AreEqual(1, clusterer.ClusterInstance(rt[0]));
Assert.AreEqual(0, clusterer.ClusterInstance(rt[2]));
Assert.AreEqual(0, clusterer.ClusterInstance(rt[3]));
Assert.AreEqual(1, clusterer.ClusterInstance(rt[4]));
}
[Test] public void test_making_single_predictions_from_trained_model()
{
var rt = Runtime.LoadFromFile <TitanicDataRow>(0, TestingHelpers.GetResourceFileName("titanic_train.csv"));
var classifier = rt.Classifiers.Trees.RandomForest.
NumExecutionSlots(4).
NumFeatures(5).
NumTrees(50);
var row = new TitanicDataRow {
age = 10,
pclass = "1",
sex = "male",
embarked = "C"
};
var prediction = classifier.ClassifyRow(row);
var proba = classifier.ClassifyRowProba(row);
Assert.AreEqual(0, prediction);
Assert.IsTrue(proba < 0.5);
}