mirror of
https://github.com/hanzoai/jube-fork.git
synced 2026-10-30 18:33:33 +00:00
richard-churchman
cc0a2bee21
Included open-source Accord.net library code natively to the solution given the deprecated state of that project. Accord.net will be supported inside Jube for own needs (such as upgrading to .Net 7+). Accord.net libraries modified to compile to .Net 6. Modified Accord.net libraries to better support Newtonsoft Json.net serialisation. Included annotations in the Accord.net libraries to house important dependency statistics. Serialised json can be used independently of Jube to recall a Neural Network. Removal of Binary Serializer totally as a precursor to the projects support of .Net 7+. Extensive bug fixing in crucial functions of Neural Network training and recall. While this part of the system remains in Alpha, it is substantially more fit for purpose.
570 lines
21 KiB
C#
570 lines
21 KiB
C#
// Accord Statistics Library
|
|
// The Accord.NET Framework
|
|
// http://accord-framework.net
|
|
//
|
|
// Copyright © César Souza, 2009-2017
|
|
// cesarsouza at gmail.com
|
|
//
|
|
// This library is free software; you can redistribute it and/or
|
|
// modify it under the terms of the GNU Lesser General Public
|
|
// License as published by the Free Software Foundation; either
|
|
// version 2.1 of the License, or (at your option) any later version.
|
|
//
|
|
// This library is distributed in the hope that it will be useful,
|
|
// but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
// Lesser General Public License for more details.
|
|
//
|
|
// You should have received a copy of the GNU Lesser General Public
|
|
// License along with this library; if not, write to the Free Software
|
|
// Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
|
|
//
|
|
#pragma warning disable 612, 618
|
|
|
|
namespace Accord.Statistics.Models.Fields.Functions
|
|
{
|
|
using System;
|
|
using System.Collections.Generic;
|
|
using Accord.Statistics.Models.Markov;
|
|
using Accord.Statistics.Models.Fields.Features;
|
|
using Accord.Statistics.Models.Fields.Functions.Specialized;
|
|
using Accord.Statistics.Distributions.Univariate;
|
|
using Accord.Math.Random;
|
|
using Accord.Compat;
|
|
|
|
/// <summary>
|
|
/// Potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
[Serializable]
|
|
public sealed class MarkovDiscreteFunction : PotentialFunctionBase<int>, IPotentialFunction<int>, ICloneable
|
|
{
|
|
|
|
/// <summary>
|
|
/// Gets the number of symbols assumed by this function.
|
|
/// </summary>
|
|
///
|
|
public int Symbols { get; private set; }
|
|
|
|
/// <summary>
|
|
/// Constructs a new potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
/// <param name="states">The number of states.</param>
|
|
/// <param name="symbols">The number of symbols.</param>
|
|
/// <param name="outputClasses">The number of output classes.</param>
|
|
///
|
|
public MarkovDiscreteFunction(int states, int symbols, int outputClasses)
|
|
{
|
|
this.Outputs = outputClasses;
|
|
this.Symbols = symbols;
|
|
|
|
int factorIndex = 0;
|
|
var factorParams = new List<double>();
|
|
var factorFeatures = new List<IFeature<int>>();
|
|
|
|
this.Factors = new FactorPotential<int>[Outputs];
|
|
|
|
int[] classOffset = new int[outputClasses];
|
|
int[] edgeOffset = new int[outputClasses];
|
|
int[] stateOffset = new int[outputClasses];
|
|
int[] classCount = new int[outputClasses];
|
|
int[] edgeCount = new int[outputClasses];
|
|
int[] stateCount = new int[outputClasses];
|
|
|
|
|
|
// Create features for initial class probabilities
|
|
for (int c = 0; c < outputClasses; c++)
|
|
{
|
|
var stateParams = new List<double>();
|
|
var stateFeatures = new List<IFeature<int>>();
|
|
|
|
var edgeParams = new List<double>();
|
|
var edgeFeatures = new List<IFeature<int>>();
|
|
|
|
var classParams = new List<double>();
|
|
var classFeatures = new List<IFeature<int>>();
|
|
|
|
|
|
// Create features for class labels
|
|
classParams.Add(Math.Log(1.0 / outputClasses));
|
|
classFeatures.Add(new OutputFeature<int>(this, c, c));
|
|
|
|
// Create features for initial state probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
edgeParams.Add((i == 0) ? Math.Log(1.0) : Math.Log(0.0));
|
|
edgeFeatures.Add(new InitialFeature<int>(this, c, i));
|
|
}
|
|
|
|
// Create features for state transition probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int j = 0; j < states; j++)
|
|
{
|
|
edgeParams.Add(Math.Log(1.0 / states));
|
|
edgeFeatures.Add(new TransitionFeature<int>(this, c, i, j));
|
|
}
|
|
}
|
|
|
|
// Create features for symbol emission probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int k = 0; k < symbols; k++)
|
|
{
|
|
stateParams.Add(Math.Log(1.0 / symbols));
|
|
stateFeatures.Add(new EmissionFeature(this, c, i, k));
|
|
}
|
|
}
|
|
|
|
classOffset[c] = factorIndex;
|
|
edgeOffset[c] = factorIndex + classParams.Count;
|
|
stateOffset[c] = factorIndex + classParams.Count + edgeParams.Count;
|
|
|
|
classCount[c] = classParams.Count;
|
|
edgeCount[c] = edgeParams.Count;
|
|
stateCount[c] = stateParams.Count;
|
|
|
|
|
|
// 1. classes
|
|
factorFeatures.AddRange(classFeatures);
|
|
factorParams.AddRange(classParams);
|
|
|
|
// 2. edges
|
|
factorFeatures.AddRange(edgeFeatures);
|
|
factorParams.AddRange(edgeParams);
|
|
|
|
// 3. states
|
|
factorFeatures.AddRange(stateFeatures);
|
|
factorParams.AddRange(stateParams);
|
|
|
|
factorIndex += classParams.Count + stateParams.Count + edgeParams.Count;
|
|
}
|
|
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorParams.Count);
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorFeatures.Count);
|
|
|
|
this.Weights = factorParams.ToArray();
|
|
this.Features = factorFeatures.ToArray();
|
|
|
|
for (int c = 0; c < outputClasses; c++)
|
|
{
|
|
Factors[c] = new MarkovDiscreteFactor(this, states, c, symbols,
|
|
classIndex: classOffset[c], classCount: classCount[c], // 1. classes
|
|
edgeIndex: edgeOffset[c], edgeCount: edgeCount[c], // 2. edges
|
|
stateIndex: stateOffset[c], stateCount: stateCount[c]); // 3. states
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Constructs a new potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
/// <param name="classifier">The classifier model.</param>
|
|
/// <param name="includePriors">True to include class features (priors), false otherwise.</param>
|
|
///
|
|
public MarkovDiscreteFunction(HiddenMarkovClassifier classifier, bool includePriors = true)
|
|
{
|
|
this.Symbols = classifier.Symbols;
|
|
this.Outputs = classifier.Classes;
|
|
|
|
int factorIndex = 0;
|
|
var factorParams = new List<double>();
|
|
var factorFeatures = new List<IFeature<int>>();
|
|
|
|
this.Factors = new FactorPotential<int>[Outputs];
|
|
|
|
int[] classOffset = new int[classifier.Classes];
|
|
int[] edgeOffset = new int[classifier.Classes];
|
|
int[] stateOffset = new int[classifier.Classes];
|
|
int[] classCount = new int[classifier.Classes];
|
|
int[] edgeCount = new int[classifier.Classes];
|
|
int[] stateCount = new int[classifier.Classes];
|
|
|
|
|
|
// Create features for initial class probabilities
|
|
for (int c = 0; c < classifier.Classes; c++)
|
|
{
|
|
var stateParams = new List<double>();
|
|
var stateFeatures = new List<IFeature<int>>();
|
|
|
|
var edgeParams = new List<double>();
|
|
var edgeFeatures = new List<IFeature<int>>();
|
|
|
|
var classParams = new List<double>();
|
|
var classFeatures = new List<IFeature<int>>();
|
|
|
|
var model = classifier[c];
|
|
|
|
if (includePriors)
|
|
{
|
|
// Create features for class labels
|
|
classParams.Add(Math.Log(classifier.Priors[c]));
|
|
classFeatures.Add(new OutputFeature<int>(this, c, c));
|
|
}
|
|
|
|
// Create features for initial state probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
edgeParams.Add(model.LogInitial[i]);
|
|
edgeFeatures.Add(new InitialFeature<int>(this, c, i));
|
|
}
|
|
|
|
// Create features for state transition probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
for (int j = 0; j < model.States; j++)
|
|
{
|
|
edgeParams.Add(model.Transitions[i, j]);
|
|
edgeFeatures.Add(new TransitionFeature<int>(this, c, i, j));
|
|
}
|
|
}
|
|
|
|
// Create features for symbol emission probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
for (int k = 0; k < model.Symbols; k++)
|
|
{
|
|
stateParams.Add(model.Emissions[i, k]);
|
|
stateFeatures.Add(new EmissionFeature(this, c, i, k));
|
|
}
|
|
}
|
|
|
|
|
|
classOffset[c] = factorIndex;
|
|
edgeOffset[c] = factorIndex + classParams.Count;
|
|
stateOffset[c] = factorIndex + classParams.Count + edgeParams.Count;
|
|
|
|
classCount[c] = classParams.Count;
|
|
edgeCount[c] = edgeParams.Count;
|
|
stateCount[c] = stateParams.Count;
|
|
|
|
|
|
// 1. classes
|
|
factorFeatures.AddRange(classFeatures);
|
|
factorParams.AddRange(classParams);
|
|
|
|
// 2. edges
|
|
factorFeatures.AddRange(edgeFeatures);
|
|
factorParams.AddRange(edgeParams);
|
|
|
|
// 3. states
|
|
factorFeatures.AddRange(stateFeatures);
|
|
factorParams.AddRange(stateParams);
|
|
|
|
factorIndex += classParams.Count + stateParams.Count + edgeParams.Count;
|
|
}
|
|
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorParams.Count);
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorFeatures.Count);
|
|
|
|
this.Weights = factorParams.ToArray();
|
|
this.Features = factorFeatures.ToArray();
|
|
|
|
|
|
for (int c = 0; c < classifier.Models.Length; c++)
|
|
{
|
|
Factors[c] = new MarkovDiscreteFactor(this, classifier.Models[c].States, c, classifier.Symbols,
|
|
classIndex: classOffset[c], classCount: classCount[c], // 1. classes
|
|
edgeIndex: edgeOffset[c], edgeCount: edgeCount[c], // 2. edges
|
|
stateIndex: stateOffset[c], stateCount: stateCount[c]); // 3. states
|
|
}
|
|
}
|
|
|
|
/// <summary>
|
|
/// Constructs a new potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
/// <param name="classifier">The classifier model.</param>
|
|
/// <param name="includePriors">True to include class features (priors), false otherwise.</param>
|
|
///
|
|
public MarkovDiscreteFunction(HiddenMarkovClassifier<GeneralDiscreteDistribution, int> classifier, bool includePriors = true)
|
|
{
|
|
this.Symbols = classifier.Models[0].Emissions[0].Length;
|
|
this.Outputs = classifier.Classes;
|
|
|
|
int factorIndex = 0;
|
|
var factorParams = new List<double>();
|
|
var factorFeatures = new List<IFeature<int>>();
|
|
|
|
this.Factors = new FactorPotential<int>[Outputs];
|
|
|
|
int[] classOffset = new int[classifier.Classes];
|
|
int[] edgeOffset = new int[classifier.Classes];
|
|
int[] stateOffset = new int[classifier.Classes];
|
|
int[] classCount = new int[classifier.Classes];
|
|
int[] edgeCount = new int[classifier.Classes];
|
|
int[] stateCount = new int[classifier.Classes];
|
|
|
|
|
|
// Create features for initial class probabilities
|
|
for (int c = 0; c < classifier.Classes; c++)
|
|
{
|
|
var stateParams = new List<double>();
|
|
var stateFeatures = new List<IFeature<int>>();
|
|
|
|
var edgeParams = new List<double>();
|
|
var edgeFeatures = new List<IFeature<int>>();
|
|
|
|
var classParams = new List<double>();
|
|
var classFeatures = new List<IFeature<int>>();
|
|
|
|
var model = classifier[c];
|
|
|
|
if (includePriors)
|
|
{
|
|
// Create features for class labels
|
|
classParams.Add(Math.Log(classifier.Priors[c]));
|
|
classFeatures.Add(new OutputFeature<int>(this, c, c));
|
|
}
|
|
|
|
// Create features for initial state probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
edgeParams.Add(model.LogInitial[i]);
|
|
edgeFeatures.Add(new InitialFeature<int>(this, c, i));
|
|
}
|
|
|
|
// Create features for state transition probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
for (int j = 0; j < model.States; j++)
|
|
{
|
|
edgeParams.Add(model.LogTransitions[i][j]);
|
|
edgeFeatures.Add(new TransitionFeature<int>(this, c, i, j));
|
|
}
|
|
}
|
|
|
|
// Create features for symbol emission probabilities
|
|
for (int i = 0; i < model.States; i++)
|
|
{
|
|
for (int k = 0; k < model.Emissions[i].Length; k++)
|
|
{
|
|
stateParams.Add(model.Emissions[i][k]);
|
|
stateFeatures.Add(new EmissionFeature(this, c, i, k));
|
|
}
|
|
}
|
|
|
|
|
|
classOffset[c] = factorIndex;
|
|
edgeOffset[c] = factorIndex + classParams.Count;
|
|
stateOffset[c] = factorIndex + classParams.Count + edgeParams.Count;
|
|
|
|
classCount[c] = classParams.Count;
|
|
edgeCount[c] = edgeParams.Count;
|
|
stateCount[c] = stateParams.Count;
|
|
|
|
|
|
// 1. classes
|
|
factorFeatures.AddRange(classFeatures);
|
|
factorParams.AddRange(classParams);
|
|
|
|
// 2. edges
|
|
factorFeatures.AddRange(edgeFeatures);
|
|
factorParams.AddRange(edgeParams);
|
|
|
|
// 3. states
|
|
factorFeatures.AddRange(stateFeatures);
|
|
factorParams.AddRange(stateParams);
|
|
|
|
factorIndex += classParams.Count + stateParams.Count + edgeParams.Count;
|
|
}
|
|
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorParams.Count);
|
|
Accord.Diagnostics.Debug.Assert(factorIndex == factorFeatures.Count);
|
|
|
|
this.Weights = factorParams.ToArray();
|
|
this.Features = factorFeatures.ToArray();
|
|
|
|
|
|
for (int c = 0; c < classifier.Models.Length; c++)
|
|
{
|
|
Factors[c] = new MarkovDiscreteFactor(this, classifier.Models[c].States, c, Symbols,
|
|
classIndex: classOffset[c], classCount: classCount[c], // 1. classes
|
|
edgeIndex: edgeOffset[c], edgeCount: edgeCount[c], // 2. edges
|
|
stateIndex: stateOffset[c], stateCount: stateCount[c]); // 3. states
|
|
}
|
|
}
|
|
|
|
|
|
/// <summary>
|
|
/// Constructs a new potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
/// <param name="states">The number of states.</param>
|
|
/// <param name="symbols">The number of symbols.</param>
|
|
/// <param name="initialization">The random number generator to use when initializing weights.</param>
|
|
///
|
|
public MarkovDiscreteFunction(int states, int symbols, IRandomNumberGenerator<double> initialization = null)
|
|
{
|
|
if (initialization == null)
|
|
initialization = new ConstantGenerator(0);
|
|
|
|
this.Symbols = symbols;
|
|
|
|
var factorParams = new List<double>();
|
|
var factorFeatures = new List<IFeature<int>>();
|
|
|
|
var stateParams = new List<double>();
|
|
var stateFeatures = new List<IFeature<int>>();
|
|
|
|
var edgeParams = new List<double>();
|
|
var edgeFeatures = new List<IFeature<int>>();
|
|
|
|
|
|
// Create features for initial state probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
edgeParams.Add(initialization.Generate());
|
|
edgeFeatures.Add(new InitialFeature<int>(this, 0, i));
|
|
}
|
|
|
|
// Create features for state transition probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int j = 0; j < states; j++)
|
|
{
|
|
edgeParams.Add(initialization.Generate());
|
|
edgeFeatures.Add(new TransitionFeature<int>(this, 0, i, j));
|
|
}
|
|
}
|
|
|
|
// Create features for symbol emission probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int k = 0; k < symbols; k++)
|
|
{
|
|
stateParams.Add(initialization.Generate());
|
|
stateFeatures.Add(new EmissionFeature(this, 0, i, k));
|
|
}
|
|
}
|
|
|
|
|
|
// 1. edges
|
|
factorFeatures.AddRange(edgeFeatures);
|
|
factorParams.AddRange(edgeParams);
|
|
|
|
// 2. states
|
|
factorFeatures.AddRange(stateFeatures);
|
|
factorParams.AddRange(stateParams);
|
|
|
|
this.Features = factorFeatures.ToArray();
|
|
this.Weights = factorParams.ToArray();
|
|
|
|
|
|
|
|
this.Factors = new[] // First features and parameters always belong to edges
|
|
{
|
|
new MarkovDiscreteFactor(this, states, 0, symbols,
|
|
edgeIndex: 0, edgeCount: edgeParams.Count, // 1. edges
|
|
stateIndex: edgeParams.Count, stateCount: stateParams.Count) // 2. states
|
|
};
|
|
}
|
|
|
|
/// <summary>
|
|
/// Constructs a new potential function modeling Hidden Markov Models.
|
|
/// </summary>
|
|
///
|
|
/// <param name="model">The hidden Markov model.</param>
|
|
///
|
|
public MarkovDiscreteFunction(HiddenMarkovModel model)
|
|
{
|
|
int states = model.States;
|
|
this.Symbols = model.Symbols;
|
|
|
|
var factorParams = new List<double>();
|
|
var factorFeatures = new List<IFeature<int>>();
|
|
|
|
var stateParams = new List<double>();
|
|
var stateFeatures = new List<IFeature<int>>();
|
|
|
|
var edgeParams = new List<double>();
|
|
var edgeFeatures = new List<IFeature<int>>();
|
|
|
|
|
|
// Create features for initial state probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
edgeParams.Add(model.LogInitial[i]);
|
|
edgeFeatures.Add(new InitialFeature<int>(this, 0, i));
|
|
}
|
|
|
|
// Create features for state transition probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int j = 0; j < states; j++)
|
|
{
|
|
edgeParams.Add(model.Transitions[i, j]);
|
|
edgeFeatures.Add(new TransitionFeature<int>(this, 0, i, j));
|
|
}
|
|
}
|
|
|
|
// Create features for symbol emission probabilities
|
|
for (int i = 0; i < states; i++)
|
|
{
|
|
for (int k = 0; k < Symbols; k++)
|
|
{
|
|
stateParams.Add(model.Emissions[i, k]);
|
|
stateFeatures.Add(new EmissionFeature(this, 0, i, k));
|
|
}
|
|
}
|
|
|
|
|
|
// 1. edges
|
|
factorFeatures.AddRange(edgeFeatures);
|
|
factorParams.AddRange(edgeParams);
|
|
|
|
// 2. states
|
|
factorFeatures.AddRange(stateFeatures);
|
|
factorParams.AddRange(stateParams);
|
|
|
|
this.Features = factorFeatures.ToArray();
|
|
this.Weights = factorParams.ToArray();
|
|
|
|
|
|
this.Factors = new[]
|
|
{
|
|
new MarkovDiscreteFactor(this, states, 0, Symbols,
|
|
edgeIndex: 0, edgeCount: edgeParams.Count, // 1. edges
|
|
stateIndex: edgeParams.Count, stateCount: stateParams.Count) // 2. states
|
|
};
|
|
}
|
|
|
|
|
|
|
|
#region ICloneable Members
|
|
|
|
private MarkovDiscreteFunction() { }
|
|
|
|
/// <summary>
|
|
/// Creates a new object that is a copy of the current instance.
|
|
/// </summary>
|
|
///
|
|
/// <returns>
|
|
/// A new object that is a copy of this instance.
|
|
/// </returns>
|
|
///
|
|
public object Clone()
|
|
{
|
|
var clone = new MarkovDiscreteFunction();
|
|
|
|
clone.Factors = new FactorPotential<int>[Factors.Length];
|
|
for (int i = 0; i < Factors.Length; i++)
|
|
clone.Factors[i] = Factors[i].Clone(newOwner: clone);
|
|
|
|
clone.Features = new IFeature<int>[Features.Length];
|
|
for (int i = 0; i < Features.Length; i++)
|
|
clone.Features[i] = Features[i].Clone(newOwner: clone);
|
|
|
|
clone.Outputs = Outputs;
|
|
clone.Symbols = Symbols;
|
|
|
|
clone.Weights = (double[])Weights.Clone();
|
|
|
|
return clone;
|
|
}
|
|
|
|
#endregion
|
|
}
|
|
}
|