/// <summary>
/// Deserializes the object from the stream.
/// </summary>
/// <param name="reader">Stream to read from.</param>
/// <returns>Node object.</returns>
public override object Read(JsonReader reader)
{
if (reader.IsNull())
{
return(null);
}
else
{
var node = (RecurrentNeuron)this.Create();
node.Label = reader.ReadProperty().Value.ToString();
node.Id = (int)reader.ReadProperty().Value;
node.NodeId = (int)reader.ReadProperty().Value;
node.LayerId = (int)reader.ReadProperty().Value;
node.IsBias = (bool)reader.ReadProperty().Value;
var activation = reader.ReadProperty().Value;
if (activation != null)
{
node.ActivationFunction = Ject.FindType(activation.ToString()).CreateDefault <IFunction>();
}
var output = reader.ReadProperty().Value;
if (output != null)
{
node.ActivationFunction = Ject.FindType(output.ToString()).CreateDefault <IFunction>();
}
node.Constrained = (bool)reader.ReadProperty().Value;
node.delta = (double)reader.ReadProperty().Value;
node.Delta = (double)reader.ReadProperty().Value;
node.Input = (double)reader.ReadProperty().Value;
node.Output = (double)reader.ReadProperty().Value;
//TODO: Read in RNN properties
return(node);
}
}
public void Test_Univariate_Back_Conversions()
{
// boolean
Assert.Equal(true, Ject.Convert(1, typeof(bool)));
Assert.Equal(false, Ject.Convert(0, typeof(bool)));
Assert.Equal(false, Ject.Convert(-1, typeof(bool)));
// numeric types
Assert.Equal((Byte)1, Ject.Convert(1d, typeof(Byte))); // byte
Assert.Equal((SByte)1, Ject.Convert(1d, typeof(SByte))); // sbyte
Assert.Equal((Decimal)0.4m, Ject.Convert(0.4d, typeof(Decimal))); // decimal
Assert.Equal((Double)0.1d, Ject.Convert(0.1d, typeof(Double))); // double
Assert.Equal((Single)300f, Ject.Convert(300d, typeof(Single))); // single
Assert.Equal((Int16)1, Ject.Convert(1d, typeof(Int16))); // int16
Assert.Equal((UInt16)2, Ject.Convert(2d, typeof(UInt16))); // uint16
Assert.Equal((Int32)2323432, Ject.Convert(2323432, typeof(Int32))); // int32
Assert.Equal((UInt32)2323432, Ject.Convert(2323432d, typeof(UInt32))); // uint32
Assert.Equal((Int64)1232323434345, Ject.Convert(1232323434345d, typeof(Int64))); // int64
Assert.Equal((UInt64)1232323434345, Ject.Convert(1232323434345d, typeof(UInt64))); // uint64
Assert.Equal((long)1232323434345, Ject.Convert(1232323434345d, typeof(long))); // long
// enum
Assert.Equal(FakeEnum.Item0, Ject.Convert(0d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item1, Ject.Convert(1d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item2, Ject.Convert(2d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item3, Ject.Convert(3d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item4, Ject.Convert(4d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item5, Ject.Convert(5d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item6, Ject.Convert(6d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item7, Ject.Convert(7d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item8, Ject.Convert(8d, typeof(FakeEnum)));
Assert.Equal(FakeEnum.Item9, Ject.Convert(9d, typeof(FakeEnum)));
// char
Assert.Equal('A', Ject.Convert(65d, typeof(char)));
// timespan
Assert.Equal(TimeSpan.FromSeconds(300), Ject.Convert(300d, typeof(TimeSpan)));
}
/// <summary>
/// Deserializes a generic model object from the stream.
/// </summary>
/// <param name="reader">A JSON Reader.</param>
/// <returns></returns>
public override object Read(JsonReader reader)
{
if (reader.IsNull())
{
return(null);
}
var model = (numl.Supervised.Model)Create();
model.Descriptor = (Descriptor)reader.ReadProperty().Value;
model.NormalizeFeatures = (bool)reader.ReadProperty().Value;
var normalizer = Ject.FindType(reader.ReadProperty().Value.ToString());
if (normalizer != null)
{
model.FeatureNormalizer = (INormalizer)Activator.CreateInstance(normalizer);
}
model.FeatureProperties = (Summary)reader.ReadProperty().Value;
return(model);
}
/// <summary>Gets best split.</summary>
/// <param name="x">The Matrix to process.</param>
/// <param name="y">The Vector to process.</param>
/// <param name="used">The used.</param>
/// <returns>The best split.</returns>
private Tuple <int, double, Impurity> GetBestSplit(Matrix x, Vector y, List <int> used)
{
double bestGain = -1;
var bestFeature = -1;
Impurity bestMeasure = null;
for (var i = 0; i < x.Cols; i++)
{
// already used?
if (used.Contains(i))
{
continue;
}
var measure = (Impurity)Ject.Create(ImpurityType);
// get appropriate column vector
var feature = x.Col(i);
// get appropriate feature at index i
// (important on because of multivalued
// cols)
var property = Descriptor.At(i);
// if discrete, calculate full relative gain
var gain = property.Discrete ? measure.RelativeGain(y, feature) : measure.SegmentedRelativeGain(y, feature, Width);
// best one?
if (!(gain > bestGain))
{
continue;
}
bestGain = gain;
bestFeature = i;
bestMeasure = measure;
}
return(new Tuple <int, double, Impurity>(bestFeature, bestGain, bestMeasure));
}
/// <summary>
/// Reads serialized neural network model
/// </summary>
/// <param name="reader">JsonReader</param>
/// <returns>Deserialized neural network model</returns>
public override object Read(JsonReader reader)
{
if (reader.IsNull())
{
return(null);
}
else
{
var model = base.Read(reader) as NeuralNetworkModel;
var outputFunction = reader.ReadProperty().Value;
if (outputFunction != null)
{
var type = Ject.FindType(outputFunction.ToString());
model.OutputFunction = (IFunction)Activator.CreateInstance(type);
}
model.Network = reader.ReadProperty().Value as Network;
return(model);
}
}
public void Test_Univariate_Conversions()
{
// boolean
Assert.AreEqual(1, Ject.Convert(true));
Assert.AreEqual(-1, Ject.Convert(false));
// numeric types
Assert.AreEqual(1d, Ject.Convert((Byte)1)); // byte
Assert.AreEqual(1d, Ject.Convert((SByte)1)); // sbyte
Assert.AreEqual(0.4d, Ject.Convert((Decimal)0.4m)); // decimal
Assert.AreEqual(0.1d, Ject.Convert((Double)0.1d)); // double
Assert.AreEqual(300d, Ject.Convert((Single)300f)); // single
Assert.AreEqual(1d, Ject.Convert((Int16)1)); // int16
Assert.AreEqual(2d, Ject.Convert((UInt16)2)); // uint16
Assert.AreEqual(2323432, Ject.Convert((Int32)2323432)); // int32
Assert.AreEqual(2323432d, Ject.Convert((UInt32)2323432)); // uint32
Assert.AreEqual(1232323434345d, Ject.Convert((Int64)1232323434345)); // int64
Assert.AreEqual(1232323434345d, Ject.Convert((UInt64)1232323434345)); // uint64
Assert.AreEqual(1232323434345d, Ject.Convert((long)1232323434345)); // long
// enum
Assert.AreEqual(0d, Ject.Convert(FakeEnum.Item0));
Assert.AreEqual(1d, Ject.Convert(FakeEnum.Item1));
Assert.AreEqual(2d, Ject.Convert(FakeEnum.Item2));
Assert.AreEqual(3d, Ject.Convert(FakeEnum.Item3));
Assert.AreEqual(4d, Ject.Convert(FakeEnum.Item4));
Assert.AreEqual(5d, Ject.Convert(FakeEnum.Item5));
Assert.AreEqual(6d, Ject.Convert(FakeEnum.Item6));
Assert.AreEqual(7d, Ject.Convert(FakeEnum.Item7));
Assert.AreEqual(8d, Ject.Convert(FakeEnum.Item8));
Assert.AreEqual(9d, Ject.Convert(FakeEnum.Item9));
// char
Assert.AreEqual(65d, Ject.Convert('A'));
// timespan
Assert.AreEqual(300d, Ject.Convert(TimeSpan.FromSeconds(300)));
}
/// <summary>Preprocess data set to create dictionary.</summary>
/// <param name="examples">.</param>
public override void PreProcess(IEnumerable <object> examples)
{
var q = from s in examples
select Ject.Get(s, Name).ToString();
if (AsEnum)
{
Dictionary = StringHelpers.BuildEnumDictionary(q).Select(kv => kv.Key).ToArray();
}
else
{
switch (SplitType)
{
case StringSplitType.Character:
Dictionary = StringHelpers.BuildCharDictionary(q, Exclude).Select(kv => kv.Key).ToArray();
break;
case StringSplitType.Word:
Dictionary = StringHelpers.BuildWordDictionary(q, Separator, Exclude).Select(kv => kv.Key).ToArray();
break;
}
}
}
public override object Read(JsonReader reader)
{
var name = reader.ReadProperty().Value.ToString();
Type type = null;
var typeName = reader.ReadProperty().Value;
if (typeName != null)
{
type = Ject.FindType(typeName.ToString());
}
var start = int.Parse(reader.ReadProperty().Value.ToString());
var discrete = (bool)reader.ReadProperty().Value;
var p = (Property)Create();
p.Name = name;
p.Type = type;
p.Start = start;
p.Discrete = discrete;
return(p);
}
public void Remove(string path)
{
Ject.RemoveAt(Ject.FindIndex(x => x.Contains(path)));
}
/// <summary>
/// Writes a simple type to the underlying stream (see <seealso cref="TypeHelpers.IsSimpleType(Type, Type[])"/>)
/// </summary>
/// <param name="value"></param>
private void WriteSimpleType(object value)
{
_writer.Write(Ject.Convert(value).ToString("r", CultureInfo.InvariantCulture));
}
/// <summary>Convert the numeric representation back to the original type.</summary>
/// <param name="val">.</param>
/// <returns>An object.</returns>
public virtual object Convert(double val)
{
return(Ject.Convert(val, this.Type));
}
/// <summary>Preprocess data set to create category list.</summary>
/// <param name="examples">.</param>
public override void PreProcess(IEnumerable <object> examples)
{
var q = examples.Select(s => (Guid)Ject.Get(s, Name)).Distinct();
Categories = q.ToArray();
}
/// <summary>
/// Returns the raw value from the object for the supplied property.
/// </summary>
/// <param name="o">Object to process.</param>
/// <param name="property">Property value of the object to return.</param>
/// <returns></returns>
public object GetValue(object o, Property property)
{
return(Ject.Get(o, property.Name));
}
/// <summary>Generates a model.</summary>
/// <param name="generator">Model generator used.</param>
/// <param name="x">The Matrix to process.</param>
/// <param name="y">The Vector to process.</param>
/// <param name="examples">Source data.</param>
/// <param name="trainingPct">The training pct.</param>
/// <param name="total">Number of Examples</param>
/// <returns>The model.</returns>
private static LearningModel GenerateModel(
IGenerator generator,
Matrix x,
Vector y,
IEnumerable <object> examples,
double trainingPct,
int total)
{
var descriptor = generator.Descriptor;
//var total = examples.Count();
var trainingCount = (int)System.Math.Floor(total * trainingPct);
// 100 - trainingPercentage for testing
var testingSlice = GetTestPoints(total - trainingCount, total).ToArray();
// trainingPercentage for training
var trainingSlice = GetTrainingPoints(testingSlice, total).ToArray();
// training
var x_t = x.Slice(trainingSlice);
var y_t = y.Slice(trainingSlice);
// generate model
var model = generator.Generate(x_t, y_t);
model.Descriptor = descriptor;
var score = new Score();
if (testingSlice.Count() > 0)
{
// testing
var test = GetTestExamples(testingSlice, examples);
var y_pred = new Vector(test.Length);
var y_test = descriptor.ToExamples(test).Item2;
var isBinary = y_test.IsBinary();
if (isBinary)
{
y_test = y_test.ToBinary(f => f == 1d, 1.0, 0.0);
}
for (var j = 0; j < test.Length; j++)
{
// items under test
var o = test[j];
// make prediction
var features = descriptor.Convert(o, false).ToVector();
// --- temp changes ---
var val = model.Predict(features);
var pred = descriptor.Label.Convert(val);
var truth = Ject.Get(o, descriptor.Label.Name);
if (truth.Equals(pred))
{
y_pred[j] = y_test[j];
}
else
{
y_pred[j] = isBinary ? (y_test[j] >= 1d ? 0d : 1d) : val;
}
}
// score predictions
score = Score.ScorePredictions(y_pred, y_test);
}
return(new LearningModel {
Generator = generator, Model = model, Score = score
});
}
/// <summary>Generates a model.</summary>
/// <param name="generator">Model generator used.</param>
/// <param name="x">The Matrix to process.</param>
/// <param name="y">The Vector to process.</param>
/// <param name="examples">Source data.</param>
/// <param name="trainingPct">The training pct.</param>
/// <param name="total">Number of Examples</param>
/// <returns>The model.</returns>
private static LearningModel GenerateModel(
IGenerator generator,
Matrix x,
Vector y,
IEnumerable <object> examples,
double trainingPct,
int total)
{
var descriptor = generator.Descriptor;
// var total = examples.Count();
var trainingCount = (int)System.Math.Floor(total * trainingPct);
// 100 - trainingPercentage for testing
var testingSlice = GetTestPoints(total - trainingCount, total).ToArray();
// trainingPercentage for training
var trainingSlice = GetTrainingPoints(testingSlice, total).ToArray();
// training
var x_t = x.Slice(trainingSlice);
var y_t = y.Slice(trainingSlice);
// generate model
var model = generator.Generate(x_t, y_t);
model.Descriptor = descriptor;
// testing
var test = GetTestExamples(testingSlice, examples);
double accuracy = 0;
for (var j = 0; j < test.Length; j++)
{
// items under test
var o = test[j];
// get truth
var truth = Ject.Get(o, descriptor.Label.Name);
// if truth is a string, sanitize
if (descriptor.Label.Type == typeof(string))
{
truth = StringHelpers.Sanitize(truth.ToString());
}
// make prediction
var features = descriptor.Convert(o, false).ToVector();
var p = model.Predict(features);
var pred = descriptor.Label.Convert(p);
// assess accuracy
if (truth.Equals(pred))
{
accuracy += 1;
}
}
// get percentage correct
accuracy /= test.Length;
return(new LearningModel {
Generator = generator, Model = model, Accuracy = accuracy
});
}
public string[] GetData()
{
return(Ject.Select(x => x.GetLastPart()).ToArray());
}
public void Add(string path)
{
Ject.Add(path);
CurrentIndex = Ject.FindIndex(x => x == path);
}
