public override void Learn(SequenceData trainingData, SequenceData validationData, SequenceData testData) {
HMMGraph graph = new HMMGraph(trainingData.NumSymbols);
//Add nodes and set initial and emission probabilities
for (int i = 0; i < states; i++) {
Node new_node = new Node();
for (int s = 0; s < trainingData.NumSymbols; s++)
new_node.SetEmission(s, ran.NextDouble());
new_node.InitialProbability = ran.NextDouble();
graph.AddNode(new_node);
}
//Add random transmissions. Each node will have at most Log(n) edges in both directions
//for (int i = 0; i < graph.Nodes.Count; i++)
//{
// List<Node> shuffled = graph.Nodes.Select(e => e).ToList();
// Utilities.Shuffle(shuffled);
// int upperBound = (int)Math.Ceiling(Math.Log(graph.Nodes.Count));
// if (upperBound >= graph.Nodes.Count)
// upperBound = graph.Nodes.Count - 1;
// for (int p = 0; p <= upperBound; p++)
// {
// Node from = graph.Nodes[i];
// Node to = graph.Nodes[p];
// from.SetTransition(to, ran.NextDouble());
// }
//}
int numberOfTransitions = (int)Math.Ceiling(Math.Log(states));
foreach (Node node in graph.Nodes)
{
for (int i = 0; i < numberOfTransitions; i++)
{
Node target;
while (node.Transitions.ContainsKey(target = graph.Nodes[ran.Next(states)]));
node.SetTransition(target, ran.NextDouble());
}
}
graph.Normalize();
hmm = SparseHiddenMarkovModel.FromGraph(graph);
hmm.Learn(trainingData.GetNonempty(), tolerance);
}
private HMMGraph RandomGraph(int num_symbols)
{
HMMGraph graph = new HMMGraph(num_symbols);
graph.Nodes = Enumerable.Range(0, num_symbols).Select(_ => new Node()).ToList();
foreach (Node node in graph.Nodes)
{
node.InitialProbability = random.NextDouble();
foreach (Node n in graph.Nodes)
{
node.SetTransition(n, random.NextDouble());
}
for (int i = 0; i < num_symbols; i++)
{
node.SetEmission(i, random.NextDouble());
}
}
return graph;
}
private void SplitWorstPerformingNode(HMMGraph graph, int node)
{
Node splitNode = graph.Nodes[node];
Node newNode = new Node();
graph.AddNode(newNode);
//splitNode.InitialProbability /= 2;
newNode.InitialProbability = splitNode.InitialProbability;
foreach (Node n in splitNode.Transitions.Keys)
{
newNode.SetTransition(n, splitNode.Transitions[n]);
}
foreach (int i in splitNode.Emissions.Keys)
{
newNode.SetEmission(i, splitNode.Emissions[i]);
}
foreach (Node n in graph.Nodes)
{
if (n.Transitions.ContainsKey(splitNode))
{
//n.Transitions[splitNode] /= 2;
n.SetTransition(newNode, n.Transitions[splitNode]);
Node weakestTransition = n.Transitions.Keys.Aggregate((a, b) => ((n.Transitions[b] < n.Transitions[a]) ? b : a));
n.Transitions.Remove(weakestTransition);
}
}
graph.Normalize();
}
private int RemoveDeadStates(HMMGraph graph)
{
int deadNodes = 0;
Dictionary<Node, bool> hasInTrans = graph.Nodes.ToDictionary(node => node, _ => false);
foreach (Node node in graph.Nodes)
{
if (node.InitialProbability > 0)
{
hasInTrans[node] = true;
}
foreach (Node trans in node.Transitions.Keys)
{
hasInTrans[trans] = true;
}
}
for (int i = 0; i < graph.NumNodes; i++)
{
if (!hasInTrans[graph.Nodes[i]])
{
deadNodes++;
graph.Nodes.RemoveAt(i);
i--;
continue;
}
if (graph.Nodes[i].Transitions.Count == 0)
{
foreach (Node node in graph.Nodes)
{
node.Transitions.Remove(node);
}
deadNodes++;
graph.Nodes.RemoveAt(i);
i--;
}
}
return deadNodes;
}
private int RemoveUnpopularStates(HMMGraph graph, double[] scores)
{
int[] unpopularStates = Enumerable.Range(0, NumberOfStates).Where(i => (scores[i] == 0)).OrderByDescending(i => i).ToArray();
foreach (Node node in graph.Nodes)
{
foreach (int unpopularState in unpopularStates)
{
node.Transitions.Remove(graph.Nodes[unpopularState]);
}
}
foreach (int unpopularState in unpopularStates)
{
graph.Nodes.RemoveAt(unpopularState);
}
return unpopularStates.Length;
}
private void CutEdges(HMMGraph graph, double epsilon)
{
foreach (Node node in graph.Nodes)
{
if (node.InitialProbability < epsilon)
{
node.InitialProbability = 0;
}
node.Transitions = node.Transitions.Keys.Where(n => (node.Transitions[n] >= epsilon)).ToDictionary(n => n, n => node.Transitions[n]);
}
graph.Normalize();
}
private void SplitState(HMMGraph graph, int state)
{
Node newNode = new Node();
foreach (Node node in graph.Nodes)
{
if (node.Transitions.ContainsKey(graph.Nodes[state]))
{
//node.SetTransition(newNode, (node.Transitions[graph.Nodes[state]] / 2));
//node.Transitions[graph.Nodes[state]] /= 2;
node.SetTransition(newNode, random.NextDouble());
}
}
foreach (Node node in graph.Nodes[state].Transitions.Keys)
{
//newNode.SetTransition(node, graph.Nodes[state].Transitions[node]);
newNode.SetTransition(node, random.NextDouble());
}
foreach (int symbol in graph.Nodes[state].Emissions.Keys)
{
//newNode.SetEmission(symbol, graph.Nodes[state].Emissions[symbol]);
newNode.SetEmission(symbol, random.NextDouble());
}
//newNode.InitialProbability = (graph.Nodes[state].InitialProbability / 2);
//graph.Nodes[state].InitialProbability /= 2;
newNode.InitialProbability = random.NextDouble();
graph.Nodes.Add(newNode);
graph.Normalize();
}
private void CutEdges(HMMGraph graph, double temperature)
{
foreach (Node node in graph.Nodes)
{
node.Transitions = node.Transitions.Keys.Where(n => (node.Transitions[n] >= temperature)).ToDictionary(n => n, n => node.Transitions[n]);
}
graph.Normalize();
}