public void Test_Vector_Conversion_Simple_Numbers_And_Chars_As_Enum()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildCharDictionary(ShortStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property {
Name = "Age"
},
new Property {
Name = "Height"
},
new StringProperty {
Name = "Chars",
Dictionary = dictionary,
SplitType = StringSplitType.Character,
AsEnum = true
},
new Property {
Name = "Weight"
},
new Property {
Name = "Good"
},
};
// ["O", "N", "E", "#SYM#", "T", "W", "#NUM#", "H", "R"]
// [ 2, 1, 0, 1, 0, 0, 1, 0, 3]
var o = new { Age = 23, Height = 6.21d, Chars = "N", Weight = 220m, Good = false, };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN CHARS */
1,
/* END CHARS */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
o = new { Age = 23, Height = 6.21d, Chars = "!", Weight = 220m, Good = false, };
// array generated by descriptor ordering
truths = new double[] { 23, 6.21,
/* BEGIN CHARS */
3,
/* END CHARS */
220, -1 };
actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Strings_As_Enum()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildWordDictionary(WordStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property {
Name = "Age"
},
new Property {
Name = "Height"
},
new StringProperty {
Name = "Words", Dictionary = dictionary, AsEnum = true
},
new Property {
Name = "Weight"
},
new Property {
Name = "Good"
},
};
// [THE, QUICK, BROWN, FOX, #NUM#, SUPER, BEAR, UGLY]
// [ 0, 1, 2, 3, 4, 5, 6, 7]
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "QUICK" };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
1,
/* END TEXT */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "sUpEr" };
// array generated by descriptor ordering
truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
5,
/* END TEXT */
220, -1 };
actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
/// <summary>Generates.</summary>
/// <exception cref="InvalidOperationException">Thrown when the requested operation is invalid.</exception>
/// <param name="examples">The examples.</param>
/// <param name="k">The int to process.</param>
/// <param name="metric">(Optional) the metric.</param>
/// <returns>An int[].</returns>
public int[] Generate(IEnumerable <object> examples, int k, IDistance metric = null)
{
#region Sanity Checks
if (examples == null)
{
throw new InvalidOperationException("Cannot generate a model will no data!");
}
if (k < 2)
{
throw new InvalidOperationException("Can only cluter with k > 1");
}
if (Descriptor == null)
{
throw new InvalidOperationException("Invalid Description!");
}
var count = examples.Count();
if (k >= count)
{
throw new InvalidOperationException(
string.Format("Cannot cluster {0} items {1} different ways!", count, k));
}
#endregion
var X = Descriptor.Convert(examples).ToMatrix();
var data = Generate(X, k, metric);
return(data);
}
/// <summary>Generates.</summary>
/// <param name="descriptor">The descriptor.</param>
/// <param name="examples">The examples.</param>
public void Generate(Descriptor descriptor, IEnumerable <object> examples)
{
// generate data matrix
var x = descriptor.Convert(examples).ToMatrix();
this.Generate(x);
}
/// <summary>Generate model based on a set of examples.</summary>
/// <exception cref="InvalidOperationException">Thrown when the requested operation is invalid.</exception>
/// <param name="description">The description.</param>
/// <param name="examples">Example set.</param>
/// <returns>Model.</returns>
public IModel Generate(Descriptor description, IEnumerable <object> examples)
{
if (!examples.Any())
{
throw new InvalidOperationException("Empty example set.");
}
Descriptor = description;
if (Descriptor.Features == null || Descriptor.Features.Length == 0)
{
throw new InvalidOperationException("Invalid descriptor: Empty feature set!");
}
if (Descriptor.Label == null)
{
throw new InvalidOperationException("Invalid descriptor: Empty label!");
}
var dataset = (this.PreserveOrder ? examples : examples.Shuffle());
var doubles = Descriptor.Convert(dataset);
var(X, Y) = doubles.ToExamples();
return(Generate(X, Y));
}
public void Test_Vector_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property {
Name = "Age", Type = typeof(int)
},
new Property {
Name = "Height", Type = typeof(double)
},
new Property {
Name = "Weight", Type = typeof(decimal)
},
new Property {
Name = "Good", Type = typeof(bool)
},
};
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false };
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Dates_3()
{
Descriptor d = Generate();
d.Features[2] = new DateTimeProperty(DatePortion.Date)
{
Name = "BirthDate"
};
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, BirthDate = new DateTime(2012, 11, 7, 2, 3, 4) };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* DATE */
2012, 11, 7,
/* END DATE */
220, -1 };
var actual = d.Convert(o);
Assert.Equal(truths, actual);
// offset test
Assert.Equal(2, d.Features[2].Start);
Assert.Equal(5, d.Features[3].Start);
Assert.Equal(6, d.Features[4].Start);
// proper length
Assert.Equal(3, d.Features[2].Length);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Enumerable_Long_By_One()
{
Descriptor d = Generate();
d.Features[2] = new EnumerableProperty(10)
{
Name = "Stuff"
};
var array = new int[] { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 };
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Stuff = array };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* ARRAY */
1, 2, 3, 4, 5, 6, 7, 8, 9, 10,
/* END ARRAY */
220, -1 };
var actual = d.Convert(o);
Assert.Equal(truths, actual);
// offset test
Assert.Equal(2, d.Features[2].Start);
Assert.Equal(12, d.Features[3].Start);
Assert.Equal(13, d.Features[4].Start);
// proper length
Assert.Equal(10, d.Features[2].Length);
}
public void Test_Vector_Dictionary_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property {
Name = "Age",
},
new Property {
Name = "Height",
},
new Property {
Name = "Weight",
},
new Property {
Name = "Good",
},
};
Dictionary <string, object> item = new Dictionary <string, object>();
item["Age"] = 23;
item["Height"] = 6.21;
item["Weight"] = 220m;
item["Good"] = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(item);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Expando_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property {
Name = "Age",
},
new Property {
Name = "Height",
},
new Property {
Name = "Weight",
},
new Property {
Name = "Good",
},
};
dynamic item = new ExpandoObject();
item.Age = 23;
item.Height = 6.21;
item.Weight = 220m;
item.Good = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(item);
Assert.AreEqual(truths, actual);
}
/// <summary>
/// Generates a <see cref="IReinforcementModel"/> based on a set of temporal/continuous examples.
/// </summary>
/// <exception cref="InvalidOperationException">Thrown when the requested operation is invalid.</exception>
/// <param name="description">The description.</param>
/// <param name="examples">Example set.</param>
/// <returns>IReinforcementModel.</returns>
public IReinforcementModel Generate(Descriptor description, IEnumerable <object> examples)
{
if (examples.Count() == 0)
{
throw new InvalidOperationException("Empty example set.");
}
Descriptor = description;
if (Descriptor.Features == null || Descriptor.Features.Length == 0)
{
throw new InvalidOperationException("Invalid descriptor: Empty feature set!");
}
if (Descriptor.Label == null)
{
throw new InvalidOperationException("Invalid descriptor: Empty label!");
}
var doubles = Descriptor.Convert(examples);
var(states, actions) = doubles.ToExamples();
Vector rewards = Vector.Rand(actions.Length);
var rewardProp = description.Features.GetPropertyOfType <RewardAttribute>();
if (rewardProp != null)
{
for (int x = 0; x < examples.Count(); x++)
{
rewards[x] = rewardProp.Convert(examples.ElementAt(x)).First();
}
}
return(this.Generate(states, actions, rewards));
}
public void Test_Vector_Conversion_Simple_Numbers_And_Dates_6()
{
Descriptor d = Generate();
d.Features[2] = new DateTimeProperty(DatePortion.DateExtended | DatePortion.TimeExtended)
{
Name = "BirthDate"
};
DateTime date = new DateTime(2012, 11, 7, 2, 3, 4, 234);
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, BirthDate = date };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* DATE */
date.DayOfYear, (int)date.DayOfWeek, 4, 234,
/* END DATE */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
// offset test
Assert.AreEqual(2, d.Features[2].Start);
Assert.AreEqual(6, d.Features[3].Start);
Assert.AreEqual(7, d.Features[4].Start);
// proper length
Assert.AreEqual(4, d.Features[2].Length);
}
/// <summary>Generates.</summary>
/// <param name="desc">The description.</param>
/// <param name="examples">The examples.</param>
/// <param name="linker">The linker.</param>
/// <returns>A Cluster.</returns>
public Cluster Generate(Descriptor desc, IEnumerable<object> examples, ILinker linker)
{
// Load data
var exampleArray = examples.ToArray();
Descriptor = desc;
Matrix X = Descriptor.Convert(examples).ToMatrix();
return GenerateClustering(X, linker, exampleArray);
}
/// <summary>Generates.</summary>
/// <param name="desc">The description.</param>
/// <param name="examples">The examples.</param>
/// <param name="linker">The linker.</param>
/// <returns>A Cluster.</returns>
public Cluster Generate(Descriptor desc, IEnumerable <object> examples, ILinker linker)
{
// Load data
var exampleArray = examples.ToArray();
Descriptor = desc;
var X = Descriptor.Convert(examples).ToMatrix();
return(GenerateClustering(X, linker, exampleArray));
}
/// <summary>
/// Reinforces the model from the new item and reward.
/// </summary>
/// <param name="state">Initial State object or features with action label.</param>
/// <param name="stateP">New State object or features with reward label.</param>
public void Learn(object state, object stateP)
{
var doubles1 = Descriptor.Convert(state, true);
var tuple1 = new double[][] { doubles1.ToArray() }.ToExamples();
var doubles2 = TransitionDescriptor.Convert(stateP, true);
var tuple2 = new double[][] { doubles2.ToArray() }.ToExamples();
this.Learn(tuple1.X[0], tuple1.Y[0], tuple2.X[0], tuple2.Y[0]);
}
public Cluster Generate(Descriptor descriptor, IEnumerable<object> examples, int k, IDistance metric = null)
{
var data = examples.ToArray();
Descriptor = descriptor;
Matrix X = Descriptor.Convert(examples).ToMatrix();
var assignments = Generate(X, k, metric);
return GenerateClustering(X, assignments, data);
}
/// <summary>
/// Reinforces the model from the new item and reward.
/// </summary>
/// <param name="state">Initial State object or features with action label.</param>
/// <param name="stateP">New State object or features with reward label.</param>
public void Learn(object state, object stateP)
{
var doubles1 = Descriptor.Convert(state, true);
var tuple1 = new[] { doubles1.ToArray() }.ToExamples();
var doubles2 = TransitionDescriptor.Convert(stateP, true);
var tuple2 = new[] { doubles2.ToArray() }.ToExamples();
Learn(tuple1.Item1[0], tuple1.Item2[0], tuple2.Item1[0], tuple2.Item2[0]);
}
/// <summary>
/// Predicts the raw label value
/// </summary>
/// <param name="o">Object to predict</param>
/// <returns>Predicted value</returns>
public object PredictValue(object o)
{
if (Descriptor.Label == null)
{
throw new InvalidOperationException("Empty label precludes prediction!");
}
var y = Descriptor.Convert(o, false).ToVector();
var val = Predict(y);
var result = Descriptor.Label.Convert(val);
return(result);
}
/// <summary>Generates.</summary>
/// <param name="descriptor">The descriptor.</param>
/// <param name="examples">The examples.</param>
/// <param name="k">The int to process.</param>
/// <param name="metric">(Optional) the metric.</param>
/// <returns>An int[].</returns>
public Cluster Generate(Descriptor descriptor, IEnumerable <object> examples, int k, IDistance metric = null)
{
var data = examples.ToArray();
Descriptor = descriptor;
X = Descriptor.Convert(data).ToMatrix();
// generate assignments
var assignments = Generate(X, k, metric);
// begin packing objects into clusters
var objects = new List <object> [k];
for (var i = 0; i < assignments.Length; i++)
{
var a = assignments[i];
if (objects[a] == null)
{
objects[a] = new List <object>();
}
objects[a].Add(data[i]);
}
// create clusters
var clusters = new List <Cluster>();
for (var i = 0; i < k; i++)
{
if (!Centers[i].IsNaN()) // check for degenerate clusters
{
clusters.Add(
new Cluster
{
Id = i + 1,
Center = Centers[i],
Members = objects[i].ToArray(),
Children = new Cluster[] {}
});
}
}
// return single cluster with K children
return(new Cluster
{
Id = 0,
Center = X.Mean(VectorType.Row),
Children = clusters.ToArray()
});
}
public void Test_Vector_Conversion_OneHot_Encoding()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new OneHotProperty(4)
{
Name = "SuperOneHot"
}
};
var o = new[] {
new { SuperOneHot = "One" }, // 1, 0, 0, 0
new { SuperOneHot = "Two" }, // 0, 1, 0, 0
new { SuperOneHot = "Three" }, // 0, 0, 1, 0
new { SuperOneHot = "Two" }, // 0, 1, 0, 0
new { SuperOneHot = "Three" }, // 0, 0, 1, 0
new { SuperOneHot = "Two" }, // 0, 1, 0, 0
new { SuperOneHot = "one" }, // 1, 0, 0, 0
new { SuperOneHot = "Four" }, // 0, 0, 0, 1
new { SuperOneHot = "Two" }, // 0, 1, 0, 0
new { SuperOneHot = "one" }, // 1, 0, 0, 0
new { SuperOneHot = "Three" }, // 0, 0, 1, 0
new { SuperOneHot = "Four" } // 0, 0, 0, 1
};
var truth = new double[, ] {
{ 1, 0, 0, 0 },
{ 0, 1, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 1, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 1, 0, 0 },
{ 1, 0, 0, 0 },
{ 0, 0, 0, 1 },
{ 0, 1, 0, 0 },
{ 1, 0, 0, 0 },
{ 0, 0, 1, 0 },
{ 0, 0, 0, 1 }
}.ToTensor();
var output = d.Convert(o).ToMatrix();
Assert.Equal(truth, output);
}
public IModel Generate(Descriptor description, IEnumerable <object> examples)
{
Descriptor = description;
if (Descriptor.Features == null || Descriptor.Features.Length == 0)
{
throw new InvalidOperationException("Invalid descriptor: Empty feature set!");
}
if (Descriptor.Label == null)
{
throw new InvalidOperationException("Invalid descriptor: Empty label!");
}
var doubles = Descriptor.Convert(examples);
var tuple = doubles.ToExamples();
return(Generate(tuple.Item1, tuple.Item2));
}
public void Test_Vector_Conversion_Simple_Numbers_And_Strings()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildWordDictionary(WordStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property {
Name = "Age"
},
new Property {
Name = "Height"
},
new StringProperty {
Name = "Words", Dictionary = dictionary
},
new Property {
Name = "Weight"
},
new Property {
Name = "Good"
},
};
// [THE, QUICK, BROWN, FOX, #NUM#, SUPER, BEAR, UGLY]
// [ 1, 0, 1, 0, 1, 0, 0, 0]
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "the brown 432" };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
1, 0, 1, 0, 1, 0, 0, 0,
/* END TEXT */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
// offset test
Assert.AreEqual(10, d.Features[3].Start);
Assert.AreEqual(11, d.Features[4].Start);
}
public void Test_Matrix_Conversion_With_Label()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property {
Name = "Age"
},
new Property {
Name = "Height"
},
new Property {
Name = "Weight"
},
new Property {
Name = "Good"
},
};
d.Label = new Property {
Name = "Nice"
};
var o = new[] {
new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Nice = true },
new { Age = 12, Height = 4.2d, Weight = 120m, Good = true, Nice = false },
new { Age = 9, Height = 6.0d, Weight = 340m, Good = true, Nice = true },
new { Age = 87, Height = 3.7d, Weight = 79m, Good = false, Nice = false }
};
// array generated by descriptor ordering
// (returns IEnumerable, but should pass)
var truths = new double[][] {
new double[] { 23, 6.21, 220, -1, 1 },
new double[] { 12, 4.2, 120, 1, -1 },
new double[] { 9, 6.0, 340, 1, 1 },
new double[] { 87, 3.7, 79, -1, -1 }
};
var actual = d.Convert(o);
var array = actual.Select(s => s.ToArray()).ToArray();
Assert.Equal(truths, array);
}
public void Test_Vector_DataRow_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property {
Name = "Age",
},
new Property {
Name = "Height",
},
new Property {
Name = "Weight",
},
new Property {
Name = "Good",
},
};
DataTable table = new DataTable("student");
var age = table.Columns.Add("Age", typeof(int));
var height = table.Columns.Add("Height", typeof(double));
var weight = table.Columns.Add("Weight", typeof(decimal));
var good = table.Columns.Add("Good", typeof(bool));
DataRow row = table.NewRow();
row[age] = 23;
row[height] = 6.21;
row[weight] = 220m;
row[good] = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(row);
Assert.AreEqual(truths, actual);
}
public Cluster Generate(Descriptor descriptor, IEnumerable<object> examples, int k, IDistance metric = null)
{
var data = examples.ToArray();
Descriptor = descriptor;
X = Descriptor.Convert(data).ToMatrix();
// generate assignments
var assignments = Generate(X, k, metric);
// begin packing objects into clusters
var objects = new List<object>[k];
for (int i = 0; i < assignments.Length; i++)
{
var a = assignments[i];
if (objects[a] == null) objects[a] = new List<object>();
objects[a].Add(data[i]);
}
// create clusters
List<Cluster> clusters = new List<Cluster>();
for(int i = 0; i < k; i++)
if(!Centers[i].IsNaN()) // check for degenerate clusters
clusters.Add(new Cluster
{
Id = i + 1,
Center = Centers[i],
Members = objects[i].ToArray(),
Children = new Cluster[] { }
});
// return single cluster with K children
return new Cluster
{
Id = 0,
Center = X.Mean(VectorType.Row),
Children = clusters.ToArray()
};
}
/// <summary>
/// Generate a multi-class classification model using a specialist classifier for each class label.
/// </summary>
/// <param name="generator">The generator to use for each individual classifier.</param>
/// <param name="examples">Training examples of any number of classes</param>
/// <param name="trainingPercentage">Percentage of training examples to use, i.e. 70% = 0.7</param>
/// <param name="mixingPercentage">Percentage to mix positive and negative exmaples, i.e. 50% will add an additional 50% of
/// <paramref name="trainingPercentage"/> of negative examples into each classifier when training.</param>
/// <param name="isMultiClass">Determines whether each class is mutually inclusive.
/// <para>For example: If True, each class takes on a number of classes and does not necessarily belong to one specific class.</para>
/// <para>The ouput would then be a number of predicted classes for a single prediction. E.g. A song would be True as it may belong to classes: vocals, rock as well as bass.</para>
/// </param>
/// <returns></returns>
public static ClassificationModel Learn(IGenerator generator, IEnumerable <object> examples, double trainingPercentage, double mixingPercentage = 0.5, bool isMultiClass = true)
{
Descriptor descriptor = generator.Descriptor;
trainingPercentage = (trainingPercentage > 1.0 ? trainingPercentage / 100 : trainingPercentage);
mixingPercentage = (mixingPercentage > 1.0 ? mixingPercentage / 100 : mixingPercentage);
var classGroups = examples.Select(s => new
{
Label = generator.Descriptor.GetValue(s, descriptor.Label),
Item = s
})
.GroupBy(g => g.Label)
.ToDictionary(k => k.Key, v => v.Select(s => s.Item).ToArray());
int classes = classGroups.Count();
Dictionary <object, IClassifier> models = null;
Score finalScore = new Score();
if (classes > 2)
{
models = new Dictionary <object, IClassifier>(classes);
Task <Tuple <IClassifier, Score, object> >[] learningTasks = new Task <Tuple <IClassifier, Score, object> > [classes];
for (int y = 0; y < classes; y++)
{
models.Add(classGroups.ElementAt(y).Key, null);
int mix = (int)System.Math.Ceiling(((classGroups.ElementAt(y).Value.Count() * trainingPercentage) * mixingPercentage) / classes);
object label = classGroups.ElementAt(y).Key;
object[] truthExamples = classGroups.ElementAt(y).Value;
object[] falseExamples = classGroups.Where(w => w.Key != classGroups.Keys.ElementAt(y))
.SelectMany(s => s.Value.Take(mix).ToArray())
.ToArray();
learningTasks[y] = Task.Factory.StartNew(
() => MultiClassLearner.GenerateModel(generator, truthExamples, falseExamples, label, trainingPercentage, label)
);
}
Task.WaitAll(learningTasks);
Score[] scores = new Score[learningTasks.Count()];
for (int c = 0; c < learningTasks.Count(); c++)
{
models[learningTasks[c].Result.Item3] = learningTasks[c].Result.Item1;
scores[c] = learningTasks[c].Result.Item2;
}
finalScore = Score.CombineScores(scores);
}
else
{
// fallback to single classifier for two class classification
var dataset = descriptor.Convert(examples, true).ToExamples();
var positives = examples.Slice(dataset.Y.Indices(f => f == 1d)).ToArray();
var negatives = examples.Slice(dataset.Y.Indices(w => w != 1d)).ToArray();
var label = generator.Descriptor.GetValue(positives.First(), descriptor.Label);
var model = MultiClassLearner.GenerateModel(generator, positives, negatives, label, trainingPercentage, label);
finalScore = model.Item2;
models = new Dictionary <object, IClassifier>()
{
{ label, model.Item1 }
};
}
ClassificationModel classificationModel = new ClassificationModel()
{
Generator = generator,
Classifiers = models,
IsMultiClass = isMultiClass,
Score = finalScore
};
return(classificationModel);
}
/// <summary>
/// Generate an <see cref="IReinforcementModel" /> based on a set of examples and transitions.
/// </summary>
/// <exception cref="InvalidOperationException">Thrown when the requested operation is invalid.</exception>
/// <param name="descriptor">The description.</param>
/// <param name="examples1">Example training set or state/action pairs.</param>
/// <param name="transitionDescriptor">
/// (Optional) Descriptor for extracting transition state/reward information from
/// <paramref name="examples2" /> set.
/// </param>
/// <param name="examples2">
/// (Optional) Corresponding training set where each item represents a transition state from
/// <paramref name="examples1" /> and a reward label.
/// </param>
/// <returns>IReinforcementModel.</returns>
public IReinforcementModel Generate(
Descriptor descriptor,
IEnumerable <object> examples1,
Descriptor transitionDescriptor,
IEnumerable <object> examples2)
{
if (!examples1.Any())
{
throw new InvalidOperationException("Empty example set.");
}
var hasTransitionStates = examples2 != null && examples2.Any();
Descriptor = descriptor;
TransitionDescriptor = transitionDescriptor;
if (Descriptor.Features == null || Descriptor.Features.Length == 0)
{
throw new InvalidOperationException("Invalid State Descriptor: Empty feature set!");
}
if (Descriptor.Label == null)
{
throw new InvalidOperationException("Invalid State Descriptor: Empty label!");
}
if (hasTransitionStates)
{
if (TransitionDescriptor?.Features == null || TransitionDescriptor.Features.Length == 0)
{
throw new ArgumentNullException(
$"Transition Descriptor was null. A transition desciptor is required for '{nameof(examples2)}'.");
}
if (examples2.Count() != examples1.Count())
{
throw new InvalidOperationException(
$"Length of '{nameof(examples1)}' must match length of '{nameof(examples2)}'.");
}
}
var doubles = Descriptor.Convert(examples1);
var tuple = doubles.ToExamples();
var states = tuple.Item1.Copy();
var actions = tuple.Item2;
Matrix statesP;
var rewards = Vector.Rand(tuple.Item2.Length);
if (hasTransitionStates)
{
var doubles2 = TransitionDescriptor.Convert(examples2);
var tuple2 = doubles2.ToExamples();
statesP = tuple2.Item1;
rewards = tuple2.Item2;
}
else
{
statesP = new Matrix(states.Rows, states.Cols);
// assume temporal
for (var i = 0; i < states.Rows - 1; i++)
{
statesP[i, VectorType.Row] = states[i + 1, VectorType.Row];
}
}
return(Generate(states, actions, statesP, rewards));
}
public void Test_Vector_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property { Name = "Age", Type = typeof(int) },
new Property { Name = "Height", Type=typeof(double) },
new Property { Name = "Weight", Type = typeof(decimal) },
new Property { Name = "Good", Type=typeof(bool) },
};
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false };
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
/// <summary>Generates.</summary>
/// <param name="descriptor">The descriptor.</param>
/// <param name="examples">The examples.</param>
public void Generate(Descriptor descriptor, IEnumerable<object> examples)
{
// generate data matrix
var x = descriptor.Convert(examples).ToMatrix();
Generate(x);
}
public void Test_Vector_Dictionary_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property { Name = "Age", },
new Property { Name = "Height", },
new Property { Name = "Weight", },
new Property { Name = "Good", },
};
Dictionary<string, object> item = new Dictionary<string, object>();
item["Age"] = 23;
item["Height"] = 6.21;
item["Weight"] = 220m;
item["Good"] = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(item);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_DataRow_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property { Name = "Age", },
new Property { Name = "Height", },
new Property { Name = "Weight", },
new Property { Name = "Good", },
};
DataTable table = new DataTable("student");
var age = table.Columns.Add("Age", typeof(int));
var height = table.Columns.Add("Height", typeof(double));
var weight = table.Columns.Add("Weight", typeof(decimal));
var good = table.Columns.Add("Good", typeof(bool));
DataRow row = table.NewRow();
row[age] = 23;
row[height] = 6.21;
row[weight] = 220m;
row[good] = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(row);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Strings_As_Enum()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildWordDictionary(WordStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property { Name = "Age" },
new Property { Name = "Height" },
new StringProperty { Name = "Words", Dictionary = dictionary, AsEnum = true },
new Property { Name = "Weight" },
new Property { Name = "Good" },
};
// [THE, QUICK, BROWN, FOX, #NUM#, SUPER, BEAR, UGLY]
// [ 0, 1, 2, 3, 4, 5, 6, 7]
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "QUICK" };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
1,
/* END TEXT */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "sUpEr" };
// array generated by descriptor ordering
truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
5,
/* END TEXT */
220, -1 };
actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Strings()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildWordDictionary(WordStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property { Name = "Age" },
new Property { Name = "Height" },
new StringProperty { Name = "Words", Dictionary = dictionary },
new Property { Name = "Weight" },
new Property { Name = "Good" },
};
// [THE, QUICK, BROWN, FOX, #NUM#, SUPER, BEAR, UGLY]
// [ 1, 0, 1, 0, 1, 0, 0, 0]
var o = new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Words = "the brown 432" };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN TEXT */
1, 0, 1, 0, 1, 0, 0, 0,
/* END TEXT */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
// offset test
Assert.AreEqual(10, d.Features[3].Start);
Assert.AreEqual(11, d.Features[4].Start);
}
public void Test_Vector_Conversion_Simple_Numbers_And_Chars_As_Enum()
{
Descriptor d = new Descriptor();
var dictionary = StringHelpers.BuildCharDictionary(ShortStrings)
.Select(k => k.Key)
.ToArray();
d.Features = new Property[]
{
new Property { Name = "Age" },
new Property { Name = "Height" },
new StringProperty {
Name = "Chars",
Dictionary = dictionary,
SplitType = StringSplitType.Character,
AsEnum = true
},
new Property { Name = "Weight" },
new Property { Name = "Good" },
};
// ["O", "N", "E", "#SYM#", "T", "W", "#NUM#", "H", "R"]
// [ 2, 1, 0, 1, 0, 0, 1, 0, 3]
var o = new { Age = 23, Height = 6.21d, Chars = "N", Weight = 220m, Good = false, };
// array generated by descriptor ordering
var truths = new double[] { 23, 6.21,
/* BEGIN CHARS */
1,
/* END CHARS */
220, -1 };
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
o = new { Age = 23, Height = 6.21d, Chars = "!", Weight = 220m, Good = false, };
// array generated by descriptor ordering
truths = new double[] { 23, 6.21,
/* BEGIN CHARS */
3,
/* END CHARS */
220, -1 };
actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
public void Test_Vector_Expando_Conversion_Simple_Numbers()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property { Name = "Age", },
new Property { Name = "Height", },
new Property { Name = "Weight", },
new Property { Name = "Good", },
};
dynamic item = new ExpandoObject();
item.Age = 23;
item.Height = 6.21;
item.Weight = 220m;
item.Good = false;
var truths = new double[] { 23, 6.21, 220, -1 };
var actual = d.Convert(item);
Assert.AreEqual(truths, actual);
}
public void Test_Matrix_Conversion_With_Label()
{
Descriptor d = new Descriptor();
d.Features = new Property[]
{
new Property { Name = "Age" },
new Property { Name = "Height" },
new Property { Name = "Weight" },
new Property { Name = "Good" },
};
d.Label = new Property { Name = "Nice" };
var o = new[] {
new { Age = 23, Height = 6.21d, Weight = 220m, Good = false, Nice = true },
new { Age = 12, Height = 4.2d, Weight = 120m, Good = true, Nice = false },
new { Age = 9, Height = 6.0d, Weight = 340m, Good = true, Nice = true },
new { Age = 87, Height = 3.7d, Weight = 79m, Good = false, Nice = false }
};
// array generated by descriptor ordering
// (returns IEnumerable, but should pass)
var truths = new double[][] {
new double[] { 23, 6.21, 220, -1, 1 },
new double[] { 12, 4.2, 120, 1, -1 },
new double[] { 9, 6.0, 340, 1, 1 },
new double[] { 87, 3.7, 79, -1, -1 }
};
var actual = d.Convert(o);
Assert.AreEqual(truths, actual);
}
