public double getProbability(
int time,
int stateI,
int stateJ,
DenseDoubleVector observation,
double[][] forward,
double[][] backward) {
double num;
if (time == observation.getLength() - 1) {
num = forward[stateI][time] * transitionProbabilities.get(stateI, stateJ);
} else {
num =
forward[stateI][time]
* transitionProbabilities.get(stateI, stateJ)
* emissionProbabilities.get(stateJ, (int) observation.get(time + 1))
* backward[stateJ][time + 1];
}
double denom = 0;
for (int k = 0; k < numStates; k++) {
denom += (forward[k][time] * backward[k][time]);
}
return denom != 0.0d ? num / denom : 0.0d;
}
public double[][] calculateForwardProbabilities(DenseDoubleVector o) {
double[][] fwd = new double[numStates][o.getLength()];
// initialization (time 0)
for (int i = 0; i < numStates; i++) {
fwd[i][0] = initialProbability.get(i) * emissionProbabilities.get(i, ((int) o.get(0)));
}
// induction
for (int t = 0; t <= o.getLength() - 2; t++) {
for (int j = 0; j < numStates; j++) {
fwd[j][t + 1] = 0;
for (int i = 0; i < numStates; i++) {
fwd[j][t + 1] += (fwd[i][t] * transitionProbabilities.get(i, j));
}
fwd[j][t + 1] *= emissionProbabilities.get(j, (int) o.get(t + 1));
}
}
return fwd;
}
public double[][] calculateBackwardProbabilities(DenseDoubleVector o) {
double[][] bwd = new double[numStates][o.getLength()];
// initialization (time 0)
for (int i = 0; i < numStates; i++) {
bwd[i][o.getLength() - 1] = 1;
}
// induction
for (int t = o.getLength() - 2; t >= 0; t--) {
for (int i = 0; i < numStates; i++) {
bwd[i][t] = 0;
for (int j = 0; j < numStates; j++) {
bwd[i][t] +=
(bwd[j][t + 1]
* transitionProbabilities.get(i, j)
* emissionProbabilities.get(j, (int) o.get(t + 1)));
}
}
}
return bwd;
}
public HMM(
int numStates,
int numOutputStates,
DenseDoubleVector initialProbabilities,
DenseDoubleMatrix transitionProbabilities,
DenseDoubleMatrix emissionProbabilities) {
Preconditions.checkArgument(numStates > 1);
Preconditions.checkArgument(initialProbabilities.getLength() == numStates);
Preconditions.checkArgument(numOutputStates > 0);
this.initialProbability = initialProbabilities;
this.transitionProbabilities = transitionProbabilities;
this.numStates = numStates;
this.numOutputStates = numOutputStates;
this.emissionProbabilities = emissionProbabilities;
}
public static void forwardPropagate(
DoubleMatrix[] thetas, DoubleMatrix[] ax, DoubleMatrix[] zx, NetworkConfiguration conf) {
for (int i = 1; i < conf.layerSizes.length; i++) {
zx[i] = multiply(ax[i - 1], thetas[i - 1], false, true, conf);
if (i < (conf.layerSizes.length - 1)) {
ax[i] =
new DenseDoubleMatrix(
DenseDoubleVector.ones(zx[i].getRowCount()), conf.activations[i].apply(zx[i]));
if (conf.hiddenDropoutProbability > 0d) {
// compute dropout for ax[i]
dropout(conf.rnd, ax[i], conf.hiddenDropoutProbability);
}
} else {
// the output doesn't need a bias
ax[i] = conf.activations[i].apply(zx[i]);
}
}
}
private void print() {
System.out.println("States: " + numStates + " | OutputStates: " + numOutputStates);
System.out.println();
DecimalFormat fmt = new DecimalFormat();
fmt.setMinimumFractionDigits(5);
fmt.setMaximumFractionDigits(5);
for (int i = 0; i < numStates; i++) {
System.out.println("init(" + i + ") = " + fmt.format(initialProbability.get(i)));
}
System.out.println();
for (int i = 0; i < numStates; i++) {
for (int j = 0; j < numStates; j++) {
System.out.print(
"transition("
+ i
+ ","
+ j
+ ") = "
+ fmt.format(transitionProbabilities.get(i, j))
+ " ");
}
System.out.println();
}
System.out.println();
for (int i = 0; i < numStates; i++) {
for (int k = 0; k < numOutputStates; k++) {
System.out.print(
"emission("
+ i
+ ","
+ k
+ ") = "
+ fmt.format(emissionProbabilities.get(i, k))
+ " ");
}
System.out.println();
}
System.out.println();
}
/**
* Minimizes the given CostFunction with Nonlinear conjugate gradient method. <br>
* It uses the Polack-Ribiere (PR) to calculate the conjugate direction. See <br>
* {@link http://en.wikipedia.org/wiki/Nonlinear_conjugate_gradient_method} <br>
* for more information.
*
* @param f the cost function to minimize.
* @param pInput the input vector, also called starting point
* @param length the number of iterations to make
* @param verbose output the progress to STDOUT
* @return a vector containing the optimized input
*/
public static DoubleVector minimizeFunction(
CostFunction f, DoubleVector pInput, int length, boolean verbose) {
DoubleVector input = pInput;
int M = 0;
int i = 0; // zero the run length counter
int red = 1; // starting point
int ls_failed = 0; // no previous line search has failed
DenseDoubleVector fX = new DenseDoubleVector(0); // what we return as fX
// get function value and gradient
final Tuple<Double, DoubleVector> evaluateCost = f.evaluateCost(input);
double f1 = evaluateCost.getFirst();
DoubleVector df1 = evaluateCost.getSecond();
i = i + (length < 0 ? 1 : 0);
DoubleVector s = df1.multiply(-1.0d); // search direction is
// steepest
double d1 = s.multiply(-1.0d).dot(s); // this is the slope
double z1 = red / (1.0 - d1); // initial step is red/(|s|+1)
while (i < Math.abs(length)) { // while not finished
i = i + (length > 0 ? 1 : 0); // count iterations?!
// make a copy of current values
DoubleVector X0 = input.deepCopy();
double f0 = f1;
DoubleVector df0 = df1.deepCopy();
// begin line search
input = input.add(s.multiply(z1));
final Tuple<Double, DoubleVector> evaluateCost2 = f.evaluateCost(input);
double f2 = evaluateCost2.getFirst();
DoubleVector df2 = evaluateCost2.getSecond();
i = i + (length < 0 ? 1 : 0); // count epochs?!
double d2 = df2.dot(s);
// initialize point 3 equal to point 1
double f3 = f1;
double d3 = d1;
double z3 = -z1;
if (length > 0) {
M = MAX;
} else {
M = Math.min(MAX, -length - i);
}
// initialize quanteties
int success = 0;
double limit = -1;
while (true) {
while (((f2 > f1 + z1 * RHO * d1) | (d2 > -SIG * d1)) && (M > 0)) {
limit = z1; // tighten the bracket
double z2 = 0.0d;
double A = 0.0d;
double B = 0.0d;
if (f2 > f1) {
// quadratic fit
z2 = z3 - (0.5 * d3 * z3 * z3) / (d3 * z3 + f2 - f3);
} else {
A = 6 * (f2 - f3) / z3 + 3 * (d2 + d3); // cubic fit
B = 3 * (f3 - f2) - z3 * (d3 + 2 * d2);
// numerical error possible - ok!
z2 = (Math.sqrt(B * B - A * d2 * z3 * z3) - B) / A;
}
if (Double.isNaN(z2) || Double.isInfinite(z2)) {
z2 = z3 / 2.0d; // if we had a numerical problem then
// bisect
}
// don't accept too close to limits
z2 = Math.max(Math.min(z2, INT * z3), (1 - INT) * z3);
z1 = z1 + z2; // update the step
input = input.add(s.multiply(z2));
final Tuple<Double, DoubleVector> evaluateCost3 = f.evaluateCost(input);
f2 = evaluateCost3.getFirst();
df2 = evaluateCost3.getSecond();
M = M - 1;
i = i + (length < 0 ? 1 : 0); // count epochs?!
d2 = df2.dot(s);
z3 = z3 - z2; // z3 is now relative to the location of z2
}
if (f2 > f1 + z1 * RHO * d1 || d2 > -SIG * d1) {
break; // this is a failure
} else if (d2 > SIG * d1) {
success = 1;
break; // success
} else if (M == 0) {
break; // failure
}
double A = 6 * (f2 - f3) / z3 + 3 * (d2 + d3); // make cubic
// extrapolation
double B = 3 * (f3 - f2) - z3 * (d3 + 2 * d2);
double z2 = -d2 * z3 * z3 / (B + Math.sqrt(B * B - A * d2 * z3 * z3));
// num prob or wrong sign?
if (Double.isNaN(z2) || Double.isInfinite(z2) || z2 < 0)
if (limit < -0.5) { // if we have no upper limit
z2 = z1 * (EXT - 1); // the extrapolate the maximum
// amount
} else {
z2 = (limit - z1) / 2; // otherwise bisect
}
else if ((limit > -0.5) && (z2 + z1 > limit)) {
// extraplation beyond max?
z2 = (limit - z1) / 2; // bisect
} else if ((limit < -0.5) && (z2 + z1 > z1 * EXT)) {
// extrapolationbeyond limit
z2 = z1 * (EXT - 1.0); // set to extrapolation limit
} else if (z2 < -z3 * INT) {
z2 = -z3 * INT;
} else if ((limit > -0.5) && (z2 < (limit - z1) * (1.0 - INT))) {
// too close to the limit
z2 = (limit - z1) * (1.0 - INT);
}
// set point 3 equal to point 2
f3 = f2;
d3 = d2;
z3 = -z2;
z1 = z1 + z2;
// update current estimates
input = input.add(s.multiply(z2));
final Tuple<Double, DoubleVector> evaluateCost3 = f.evaluateCost(input);
f2 = evaluateCost3.getFirst();
df2 = evaluateCost3.getSecond();
M = M - 1;
i = i + (length < 0 ? 1 : 0); // count epochs?!
d2 = df2.dot(s);
} // end of line search
DoubleVector tmp = null;
if (success == 1) { // if line search succeeded
f1 = f2;
fX = new DenseDoubleVector(fX.toArray(), f1);
if (verbose) System.out.print("Iteration " + i + " | Cost: " + f1 + "\r");
// Polack-Ribiere direction: s =
// (df2'*df2-df1'*df2)/(df1'*df1)*s - df2;
final double numerator = (df2.dot(df2) - df1.dot(df2)) / df1.dot(df1);
s = s.multiply(numerator).subtract(df2);
tmp = df1;
df1 = df2;
df2 = tmp; // swap derivatives
d2 = df1.dot(s);
if (d2 > 0) { // new slope must be negative
s = df1.multiply(-1.0d); // otherwise use steepest direction
d2 = s.multiply(-1.0d).dot(s);
}
// realmin in octave = 2.2251e-308
// slope ratio but max RATIO
z1 = z1 * Math.min(RATIO, d1 / (d2 - 2.2251e-308));
d1 = d2;
ls_failed = 0; // this line search did not fail
} else {
input = X0;
f1 = f0;
df1 = df0; // restore point from before failed line search
// line search failed twice in a row?
if (ls_failed == 1 || i > Math.abs(length)) {
break; // or we ran out of time, so we give up
}
tmp = df1;
df1 = df2;
df2 = tmp; // swap derivatives
s = df1.multiply(-1.0d); // try steepest
d1 = s.multiply(-1.0d).dot(s);
z1 = 1.0d / (1.0d - d1);
ls_failed = 1; // this line search failed
}
}
return input;
}
public void trainBaumWelch(List<DenseDoubleVector> observations, int steps) {
double pi1[] = new double[numStates];
double a1[][] = new double[numStates][numStates];
double b1[][] = new double[numStates][numOutputStates];
for (int s = 0; s < steps; s++) {
for (DenseDoubleVector o : observations) {
// calculate forward and backward probabilities
double[][] fwd = calculateForwardProbabilities(o);
double[][] bwd = calculateBackwardProbabilities(o);
// re-estimation of initial state probabilities
for (int i = 0; i < numStates; i++) {
pi1[i] = gamma(i, 0, fwd, bwd);
}
// re-estimation of transition probabilities
for (int i = 0; i < numStates; i++) {
for (int j = 0; j < numStates; j++) {
double num = 0;
double denom = 0;
for (int t = 0; t <= o.getLength() - 1; t++) {
num += getProbability(t, i, j, o, fwd, bwd);
denom += gamma(i, t, fwd, bwd);
}
a1[i][j] = denom != 0.0d ? num / denom : 0.0d;
}
}
// re-estimation of emission probabilities
for (int i = 0; i < numStates; i++) {
for (int k = 0; k < numOutputStates; k++) {
double num = 0;
double denom = 0;
for (int t = 0; t <= o.getLength() - 1; t++) {
double g = gamma(i, t, fwd, bwd);
num += g * (k == o.get(t) ? 1.0d : 0.0d);
denom += g;
}
b1[i][k] = denom != 0.0d ? num / denom : 0.0d;
}
}
}
// TODO calculate the kullback leibler divergence of the output
for (int i = 0; i < initialProbability.getLength(); i++) {
initialProbability.set(i, pi1[i]);
}
for (int col = 0; col < transitionProbabilities.getColumnCount(); col++) {
for (int row = 0; row < transitionProbabilities.getRowCount(); row++) {
transitionProbabilities.set(row, col, a1[row][col]);
}
}
for (int col = 0; col < emissionProbabilities.getColumnCount(); col++) {
for (int row = 0; row < emissionProbabilities.getRowCount(); row++) {
emissionProbabilities.set(row, col, b1[row][col]);
}
}
}
}
