public Double eval(Double gamma) {
double c = 1.0 / (double) Math.sqrt(variance);
double sum = 0.0;
for (Integer i : transcripts.get(gammat)) {
double pi = 0.0;
for (int t = 0; t < fiveprime.length; t++) {
if (transcripts.get(t).contains(i)) {
double gammai = gammat == t ? gamma : gammas[t][gammak];
double dit = delta(i, t);
double pit = gammai * Math.exp(-lambda * dit);
pi += pit;
}
}
double zi = Math.log(pi);
double err = data.values(i)[channels[gammak]] - zi;
double ratio = (Math.exp(-lambda * delta(i, gammat))) / pi;
double termi = (err * ratio);
sum += termi;
}
return c * sum;
}
private void getnegphase() {
/*
* It does the negative phase of unsupervised RBM training algorithm
*
* For details, please refer to Dr. Hinton's paper:
* Reducing the dimensionality of data with neural networks. Science, Vol. 313. no. 5786, pp. 504 - 507, 28 July 2006.
*/
// start calculate the negative phase
// calculate the curved value of v1,h1
// find the vector of v1
Matrix negdata = poshidstates.times(vishid.transpose());
// (1 * numhid) * (numhid * numdims) = (1 * numdims)
negdata.plusEquals(visbiases);
// poshidstates*vishid' + visbiases
double[][] tmp1 = negdata.getArray();
int i1 = 0;
while (i1 < numdims) {
tmp1[0][i1] = 1 / (1 + Math.exp(-tmp1[0][i1]));
i1++;
}
// find the vector of h1
neghidprobs = negdata.times(vishid);
// (1 * numdims) * (numdims * numhid) = (1 * numhid)
neghidprobs.plusEquals(hidbiases);
double[][] tmp2 = neghidprobs.getArray();
int i2 = 0;
while (i2 < numhid) {
tmp2[0][i2] = 1 / (1 + Math.exp(-tmp2[0][i2]));
i2++;
}
negprods = negdata.transpose().times(neghidprobs);
// (numdims * 1) *(1 * numhid) = (numdims * numhid)
}
private ArrayList<Double> forwardProp(Double[] input) {
for (int i = 0; i < hiddenLNo; i++) {
for (int j = 0; j < inputSize + 1; j++) {
if (j == inputSize) {
hiddenNodes.set(i, hiddenNodes.get(i) + w1[j][i] * bias);
} else {
hiddenNodes.set(i, hiddenNodes.get(i) + w1[j][i] * input[j]);
}
}
double temp = (1.0 / (1.0 + Math.exp(-hiddenNodes.get(i))));
hiddenNodes.set(i, temp);
}
for (int i = 0; i < outputLNo; i++) {
for (int j = 0; j < hiddenLNo + 1; j++) {
if (j == hiddenNodes.size()) {
outputNodes.set(i, outputNodes.get(i) + w2[j][i] * bias);
} else {
outputNodes.set(i, outputNodes.get(i) + w2[j][i] * hiddenNodes.get(j));
}
}
// double temp1 = (1.0 / (1.0 + Math.exp(-outputNodes.get(i))));
outputNodes.set(i, (1.0 / (1.0 + Math.exp(-outputNodes.get(i)))));
}
return outputNodes;
}
public Double eval(Double gamma) {
double c1 = 0.0;
double c3 = -1.0 / (double) Math.sqrt(2.0 * variance);
double sum = 0.0;
double cpart = -0.5 * Math.log(Math.sqrt(2.0 * Math.PI * variance));
for (Integer i : transcripts.get(gammat)) {
c1 += cpart;
double pi = 0.0;
for (int t = 0; t < fiveprime.length; t++) {
if (transcripts.get(t).contains(i)) {
double gammai = gammat == t ? gamma : gammas[t][gammak];
double dit = delta(i, t);
double pit = gammai * Math.exp(-lambda * dit);
pi += pit;
}
}
double zi = Math.log(pi);
double err = data.values(i)[channels[gammak]] - zi;
sum += (err * err);
}
return c1 + c3 * sum;
}
public double error() {
double sum = 0.0;
for (int i = 0; i < data.size(); i++) {
for (int k = 0; k < channels.length; k++) {
double pi = 0.0;
for (Integer t : transcripts.keySet()) {
if (transcripts.get(t).contains(i)) {
double gammai = gammas[t][k];
double dit = delta(i, t);
double pit = gammai * Math.exp(-lambda * dit);
pi += pit;
}
}
double zi = Math.log(pi);
double err = Math.abs(data.values(i)[channels[k]] - zi);
sum += err;
}
}
return sum;
}
public Double eval(Double lmbda) {
double c = 1.0 / (double) Math.sqrt(variance);
double sum = 0.0;
for (int i = 0; i < data.size(); i++) {
for (int k = 0; k < channels.length; k++) {
double pi = 0.0;
double pdi = 0.0;
for (Integer t : transcripts.keySet()) {
if (transcripts.get(t).contains(i)) {
double gammai = gammas[t][k];
double dit = delta(i, t);
double falloff = Math.exp(-lmbda * dit);
double pit = gammai * falloff;
double pdit = pit * dit;
pi += pit;
pdi += pdit;
}
}
double zi = Math.log(pi);
double err = data.values(i)[channels[k]] - zi;
double ratio = pdi / pi;
double termi = (err * ratio);
sum += termi;
}
}
return c * sum;
}
private void rescale(long newLandmarkInSeconds) {
// rescale the weights based on a new landmark to avoid numerical overflow issues
final double factor = Math.exp(-alpha * (newLandmarkInSeconds - landmarkInSeconds));
weightedCount *= factor;
postOrderTraversal(
root,
new Callback() {
@Override
public boolean process(Node node) {
double oldWeight = node.weightedCount;
node.weightedCount *= factor;
if (oldWeight >= ZERO_WEIGHT_THRESHOLD && node.weightedCount < ZERO_WEIGHT_THRESHOLD) {
--nonZeroNodeCount;
}
return true;
}
});
landmarkInSeconds = newLandmarkInSeconds;
}
/**
* Apply this operator (function) to the supplied argument
*
* @param value the argument
* @return the result
*/
protected double applyFunction(double value) {
switch (m_operator) {
case 'l':
return Math.log(value);
case 'b':
return Math.abs(value);
case 'c':
return Math.cos(value);
case 'e':
return Math.exp(value);
case 's':
return Math.sqrt(value);
case 'f':
return Math.floor(value);
case 'h':
return Math.ceil(value);
case 'r':
return Math.rint(value);
case 't':
return Math.tan(value);
case 'n':
return Math.sin(value);
}
return Double.NaN;
}
public Double eval(Double lmbda) {
double c1 = 0.0;
double c3 = -1.0 / (double) Math.sqrt(2.0 * variance);
double sum = 0.0;
double cpart = -0.5 * Math.log(Math.sqrt(2.0 * Math.PI * variance));
for (int i = 0; i < data.size(); i++) {
for (int k = 0; k < channels.length; k++) {
c1 += cpart;
double pi = 0.0;
for (Integer t : transcripts.keySet()) {
if (transcripts.get(t).contains(i)) {
double dit = delta(i, t);
double gammai = gammas[t][k];
double pit = gammai * Math.exp(-lmbda * dit);
pi += pit;
}
}
double zi = Math.log(pi);
double err = data.values(i)[channels[k]] - zi;
sum += (err * err);
}
}
return c1 + c3 * sum;
}
public static double qt(double p, double ndf, boolean lower_tail) {
// Algorithm 396: Student's t-quantiles by
// G.W. Hill CACM 13(10), 619-620, October 1970
if (p <= 0 || p >= 1 || ndf < 1)
throw new IllegalArgumentException("Invalid p or df in call to qt(double,double,boolean).");
double eps = 1e-12;
double M_PI_2 = 1.570796326794896619231321691640; // pi/2
boolean neg;
double P, q, prob, a, b, c, d, y, x;
if ((lower_tail && p > 0.5) || (!lower_tail && p < 0.5)) {
neg = false;
P = 2 * (lower_tail ? (1 - p) : p);
} else {
neg = true;
P = 2 * (lower_tail ? p : (1 - p));
}
if (Math.abs(ndf - 2) < eps) {
/* df ~= 2 */
q = Math.sqrt(2 / (P * (2 - P)) - 2);
} else if (ndf < 1 + eps) {
/* df ~= 1 */
prob = P * M_PI_2;
q = Math.cos(prob) / Math.sin(prob);
} else {
/*-- usual case; including, e.g., df = 1.1 */
a = 1 / (ndf - 0.5);
b = 48 / (a * a);
c = ((20700 * a / b - 98) * a - 16) * a + 96.36;
d = ((94.5 / (b + c) - 3) / b + 1) * Math.sqrt(a * M_PI_2) * ndf;
y = Math.pow(d * P, 2 / ndf);
if (y > 0.05 + a) {
/* Asymptotic inverse expansion about normal */
x = qnorm(0.5 * P, false);
y = x * x;
if (ndf < 5) c += 0.3 * (ndf - 4.5) * (x + 0.6);
c = (((0.05 * d * x - 5) * x - 7) * x - 2) * x + b + c;
y = (((((0.4 * y + 6.3) * y + 36) * y + 94.5) / c - y - 3) / b + 1) * x;
y = a * y * y;
if (y > 0.002) /* FIXME: This cutoff is machine-precision dependent*/ y = Math.exp(y) - 1;
else {
/* Taylor of e^y -1 : */
y = (0.5 * y + 1) * y;
}
} else {
y =
((1 / (((ndf + 6) / (ndf * y) - 0.089 * d - 0.822) * (ndf + 2) * 3) + 0.5 / (ndf + 4))
* y
- 1)
* (ndf + 1)
/ (ndf + 2)
+ 1 / y;
}
q = Math.sqrt(ndf * y);
}
if (neg) q = -q;
return q;
}
public static float gaussian(float x, float mean, float sd) {
float mu = mean;
float sigma = sd;
float k1 = (float) ((float) 1 / (sigma * (Math.sqrt(2 * Math.PI))));
float k2 = -1 / (2 * (sigma * sigma));
return (float) (k1 * Math.exp((x - mu) * (x - mu) * k2));
}
public static double g(Double[] xi, int yi) {
// yi sum (k, wk xik)
double sum = weightvector[0]; // w0
for (int k = 1; k < numfeatures + 1; k++) {
sum += weightvector[k] * xi[k];
}
return (1 / (1 + Math.exp(yi * sum))); // exp(-(-yi*sum))
} // end g(z)
static double k_function(svm_node[] x, svm_node[] y,
svm_parameter param)
{
switch(param.kernel_type)
{
case svm_parameter.LINEAR:
return dot(x,y);
case svm_parameter.POLY:
return Math.pow(param.gamma*dot(x,y)+param.coef0,param.degree);
case svm_parameter.RBF:
{
double sum = 0;
int xlen = x.length;
int ylen = y.length;
int i = 0;
int j = 0;
while(i < xlen && j < ylen)
{
if(x[i].index == y[j].index)
{
double d = x[i++].value - y[j++].value;
sum += d*d;
}
else if(x[i].index > y[j].index)
{
sum += y[j].value * y[j].value;
++j;
}
else
{
sum += x[i].value * x[i].value;
++i;
}
}
while(i < xlen)
{
sum += x[i].value * x[i].value;
++i;
}
while(j < ylen)
{
sum += y[j].value * y[j].value;
++j;
}
return Math.exp(-param.gamma*sum);
}
case svm_parameter.SIGMOID:
return tanh(param.gamma*dot(x,y)+param.coef0);
default:
System.err.print("unknown kernel function.\n");
System.exit(1);
return 0; // java
}
}
public static double pnorm(double z, boolean upper) {
/* Reference:
I. D. Hill
Algorithm AS 66: "The Normal Integral"
Applied Statistics
*/
double ltone = 7.0,
utzero = 18.66,
con = 1.28,
a1 = 0.398942280444,
a2 = 0.399903438504,
a3 = 5.75885480458,
a4 = 29.8213557808,
a5 = 2.62433121679,
a6 = 48.6959930692,
a7 = 5.92885724438,
b1 = 0.398942280385,
b2 = 3.8052e-8,
b3 = 1.00000615302,
b4 = 3.98064794e-4,
b5 = 1.986153813664,
b6 = 0.151679116635,
b7 = 5.29330324926,
b8 = 4.8385912808,
b9 = 15.1508972451,
b10 = 0.742380924027,
b11 = 30.789933034,
b12 = 3.99019417011;
double y, alnorm;
if (z < 0) {
upper = !upper;
z = -z;
}
if (z <= ltone || upper && z <= utzero) {
y = 0.5 * z * z;
if (z > con) {
alnorm =
b1
* Math.exp(-y)
/ (z
- b2
+ b3
/ (z
+ b4
+ b5 / (z - b6 + b7 / (z + b8 - b9 / (z + b10 + b11 / (z + b12))))));
} else {
alnorm = 0.5 - z * (a1 - a2 * y / (y + a3 - a4 / (y + a5 + a6 / (y + a7))));
}
} else {
alnorm = 0;
}
if (!upper) alnorm = 1 - alnorm;
return (alnorm);
}
public String reportLine() {
return String.format(
"%s, pval= %.2e, %d / %d = %.3f (freq in GO %d / %d = %.3f)",
getCategory(),
Math.exp(getLogPValue()),
getx(),
getn(),
(double) ((double) getx()) / getn(),
getTheta(),
getN(),
(double) ((double) getTheta()) / getN());
}
private double computeDistortionProb(int englishPosition, int frenchPosition,
int numEnglishWords, int numFrenchWords) {
double alpha = 1.0;
double kNull = .2; // proportion of probability mass to allot for null alignments
if (englishPosition == 0) { // How to compute distortion probability for null alignments?
return kNull;
}
else {
double dist = englishPosition - (frenchPosition * numEnglishWords / numFrenchWords);
double metric = (1.0 - kNull) * Math.exp(-1.0 * alpha * dist);
return metric;
}
}
public static DComplex power(double x_re, double x_im, double y_re, double y_im) {
double h;
/* #ifdef JAVA5 */
// h = Math.hypot(x_re, x_im);
/* #else */
h = DComplex.hypot(x_re, x_im);
/* #endif */
double logr = Math.log(h);
double t = Math.atan2(x_im, x_re);
double r = Math.exp(logr * y_re - y_im * t);
t = y_im * logr + y_re * t;
return Complex.polar(r, t);
}
public double pchisq(double q, double df) {
// Posten, H. (1989) American Statistician 43 p. 261-265
double df2 = df * .5;
double q2 = q * .5;
int n = 5, k;
double tk, CFL, CFU, prob;
if (q <= 0 || df <= 0)
throw new IllegalArgumentException("Illegal argument " + q + " or " + df + " for qnorm(p).");
if (q < df) {
tk = q2 * (1 - n - df2) / (df2 + 2 * n - 1 + n * q2 / (df2 + 2 * n));
for (k = n - 1; k > 1; k--)
tk = q2 * (1 - k - df2) / (df2 + 2 * k - 1 + k * q2 / (df2 + 2 * k + tk));
CFL = 1 - q2 / (df2 + 1 + q2 / (df2 + 2 + tk));
prob = Math.exp(df2 * Math.log(q2) - q2 - Maths.logGamma(df2 + 1) - Math.log(CFL));
} else {
tk = (n - df2) / (q2 + n);
for (k = n - 1; k > 1; k--) tk = (k - df2) / (q2 + k / (1 + tk));
CFU = 1 + (1 - df2) / (q2 + 1 / (1 + tk));
prob = 1 - Math.exp((df2 - 1) * Math.log(q2) - q2 - Maths.logGamma(df2) - Math.log(CFU));
}
return prob;
}
/**
* 预测
*
* @param context 环境
* @param prior 先验概率
* @param model 特征函数
* @return
*/
public static double[] eval(int[] context, double[] prior, EvalParameters model) {
Context[] params = model.getParams();
int numfeats[] = new int[model.getNumOutcomes()];
int[] activeOutcomes;
double[] activeParameters;
double value = 1;
for (int ci = 0; ci < context.length; ci++) {
if (context[ci] >= 0) {
Context predParams = params[context[ci]];
activeOutcomes = predParams.getOutcomes();
activeParameters = predParams.getParameters();
for (int ai = 0; ai < activeOutcomes.length; ai++) {
int oid = activeOutcomes[ai];
numfeats[oid]++;
prior[oid] += activeParameters[ai] * value;
}
}
}
double normal = 0.0;
for (int oid = 0; oid < model.getNumOutcomes(); oid++) {
if (model.getCorrectionParam() != 0) {
prior[oid] =
Math.exp(
prior[oid] * model.getConstantInverse()
+ ((1.0 - ((double) numfeats[oid] / model.getCorrectionConstant()))
* model.getCorrectionParam()));
} else {
prior[oid] = Math.exp(prior[oid] * model.getConstantInverse());
}
normal += prior[oid];
}
for (int oid = 0; oid < model.getNumOutcomes(); oid++) {
prior[oid] /= normal;
}
return prior;
}
public static double betainv(double x, double p, double q) {
// ALGORITHM AS 63 APPL. STATIST. VOL.32, NO.1
// Computes P(Beta>x)
double beta = Maths.logBeta(p, q), acu = 1E-14;
double cx, psq, pp, qq, x2, term, ai, betain, ns, rx, temp;
boolean indx;
if (p <= 0 || q <= 0) return (-1.0);
if (x <= 0 || x >= 1) return (-1.0);
psq = p + q;
cx = 1 - x;
if (p < psq * x) {
x2 = cx;
cx = x;
pp = q;
qq = p;
indx = true;
} else {
x2 = x;
pp = p;
qq = q;
indx = false;
}
term = 1;
ai = 1;
betain = 1;
ns = qq + cx * psq;
rx = x2 / cx;
temp = qq - ai;
if (ns == 0) rx = x2;
while (temp > acu && temp > acu * betain) {
term = term * temp * rx / (pp + ai);
betain = betain + term;
temp = Math.abs(term);
if (temp > acu && temp > acu * betain) {
ai++;
ns--;
if (ns >= 0) {
temp = qq - ai;
if (ns == 0) rx = x2;
} else {
temp = psq;
psq += 1;
}
}
}
betain *= Math.exp(pp * Math.log(x2) + (qq - 1) * Math.log(cx) - beta) / pp;
if (indx) betain = 1 - betain;
return (betain);
}
double kernel_function(int i, int j)
{
switch(kernel_type)
{
case svm_parameter.LINEAR:
return dot(x[i],x[j]);
case svm_parameter.POLY:
return Math.pow(gamma*dot(x[i],x[j])+coef0,degree);
case svm_parameter.RBF:
return Math.exp(-gamma*(x_square[i]+x_square[j]-2*dot(x[i],x[j])));
case svm_parameter.SIGMOID:
return tanh(gamma*dot(x[i],x[j])+coef0);
default:
System.err.print("unknown kernel function.\n");
System.exit(1);
return 0; // java
}
}
private void prop2nextLayer() {
/*
* It computes the forward propagation algorithm.
*/
poshidprobs = data.times(vishid);
// (1 * numdims) * (numdims * numhid)
poshidprobs.plusEquals(hidbiases);
// data*vishid + hidbiases
double[][] product_tmp2 = poshidprobs.getArray();
for (int i2 = 0; i2 < numhid; i2++) {
/*
* compute the updated input, and write them to newinput
*/
product_tmp2[0][i2] = 1 / (1 + Math.exp(-product_tmp2[0][i2]));
newinput[i2] = (int) (product_tmp2[0][i2] * 255.0);
}
}
/** log_sum_exp: for inputs a1, a2, ... output log (e^a1 + e^a2 + ... ) */
public static double log_sum_exp(List<Double> ls) {
double ins_log = 0;
Collections.sort(ls);
double v_max = ls.get(ls.size() - 1);
if (Double.isNaN(v_max)) {
System.out.println("ERROR6");
}
for (int i = 0; i < ls.size(); i++) {
ins_log += Math.exp(ls.get(i) - v_max);
}
double res = v_max + Math.log(ins_log);
if (Double.isNaN(res)) {
System.out.println("ERROR4");
Scanner sc = new Scanner(System.in);
int gu = sc.nextInt();
}
return res;
}
/**
* Get the top few clauses from this searcher, cutting off at the given minimum probability.
*
* @param thresholdProbability The threshold under which to stop returning clauses. This should be
* between 0 and 1.
* @return The resulting {@link edu.stanford.nlp.naturalli.SentenceFragment} objects, representing
* the top clauses of the sentence.
*/
public List<SentenceFragment> topClauses(double thresholdProbability) {
List<SentenceFragment> results = new ArrayList<>();
search(
triple -> {
assert triple.first <= 0.0;
double prob = Math.exp(triple.first);
assert prob <= 1.0;
assert prob >= 0.0;
assert !Double.isNaN(prob);
if (prob >= thresholdProbability) {
SentenceFragment fragment = triple.third.get();
fragment.score = prob;
results.add(fragment);
return true;
} else {
return false;
}
});
return results;
}
private void getposphase() {
/*
* It does the positive phase of unsupervised RBM training algorithm
*
* For details, please refer to Dr. Hinton's paper:
* Reducing the dimensionality of data with neural networks. Science, Vol. 313. no. 5786, pp. 504 - 507, 28 July 2006.
*/
// Start calculate the positive phase
// calculate the cured value of h0
poshidprobs = data.times(vishid);
// (1 * numdims) * (numdims * numhid)
poshidprobs.plusEquals(hidbiases);
// data*vishid + hidbiases
double[][] product_tmp2 = poshidprobs.getArray();
int i2 = 0;
while (i2 < numhid) {
product_tmp2[0][i2] = 1 / (1 + Math.exp(-product_tmp2[0][i2]));
i2++;
}
posprods = data.transpose().times(poshidprobs);
// (numdims * 1) * (1 * numhid)
// end of the positive phase calculation, find the binary presentation of h0
int i3 = 0;
double[][] tmp1 = poshidprobs.getArray();
double[][] tmp2 = new double[1][numhid];
Random randomgenerator = new Random();
while (i3 < numhid) {
/*
* a sampling according to possiblity given by poshidprobs
*/
if (tmp1[0][i3] > randomgenerator.nextDouble()) tmp2[0][i3] = 1;
else tmp2[0][i3] = 0;
i3++;
}
// poshidstates is a binary sampling according to possiblity given by poshidprobs
poshidstates = new Matrix(tmp2);
}
// returns true if neuron spike caused by input spike at time t in us
boolean stimulate(int t) {
if (!initialized) {
lastt = t;
initialized = true;
return true;
}
if (t < lastt) {
reset();
return true;
}
int dt = t - lastt;
float delta = dt / filter.tauUs;
float exp = delta > 20 ? 0 : (float) Math.exp(-delta);
float newstate = state * exp + filter.weight;
if (newstate > max) {
newstate = max;
}
boolean spike = random.nextFloat() > state; // spike goes through based on decayed state
state = newstate;
lastt = t;
return spike;
}
private static double tanh(double x)
{
double e = Math.exp(x);
return 1.0-2.0/(e*e+1);
}
private double weight(long timestamp) {
return Math.exp(alpha * (timestamp - landmarkInSeconds));
}
/**
* Calculate the probability of a dependency as a real probability between 0 and 1 inclusive.
*
* @param dependency The dependency for which the probability is to be calculated. The tags in
* this dependency are in the reduced TagProjection space.
* @return The probability of the dependency
*/
protected double probTB(IntDependency dependency) {
if (verbose) {
// System.out.println("tagIndex: " + tagIndex);
System.err.println("Generating " + dependency);
}
boolean leftHeaded = dependency.leftHeaded && directional;
int hW = dependency.head.word;
int aW = dependency.arg.word;
short hT = dependency.head.tag;
short aT = dependency.arg.tag;
IntTaggedWord aTW = dependency.arg;
IntTaggedWord hTW = dependency.head;
boolean isRoot = rootTW(dependency.head);
double pb_stop_hTWds;
if (isRoot) {
pb_stop_hTWds = 0.0;
} else {
pb_stop_hTWds = getStopProb(dependency);
}
if (dependency.arg.word == STOP_WORD_INT) {
// did we generate stop?
return pb_stop_hTWds;
}
double pb_go_hTWds = 1.0 - pb_stop_hTWds;
// generate the argument
short binDistance = valenceBin(dependency.distance);
// KEY:
// c_ count of (read as joint count of first and second)
// p_ MLE prob of (or MAP if useSmoothTagProjection)
// pb_ MAP prob of (read as prob of first given second thing)
// a arg
// h head
// T tag
// PT projected tag
// W word
// d direction
// ds distance (implicit: there when direction is mentioned!)
IntTaggedWord anyHead = new IntTaggedWord(ANY_WORD_INT, dependency.head.tag);
IntTaggedWord anyArg = new IntTaggedWord(ANY_WORD_INT, dependency.arg.tag);
IntTaggedWord anyTagArg = new IntTaggedWord(dependency.arg.word, ANY_TAG_INT);
IntDependency temp =
new IntDependency(dependency.head, dependency.arg, leftHeaded, binDistance);
double c_aTW_hTWd = argCounter.getCount(temp);
temp = new IntDependency(dependency.head, anyArg, leftHeaded, binDistance);
double c_aT_hTWd = argCounter.getCount(temp);
temp = new IntDependency(dependency.head, wildTW, leftHeaded, binDistance);
double c_hTWd = argCounter.getCount(temp);
temp = new IntDependency(anyHead, dependency.arg, leftHeaded, binDistance);
double c_aTW_hTd = argCounter.getCount(temp);
temp = new IntDependency(anyHead, anyArg, leftHeaded, binDistance);
double c_aT_hTd = argCounter.getCount(temp);
temp = new IntDependency(anyHead, wildTW, leftHeaded, binDistance);
double c_hTd = argCounter.getCount(temp);
// for smooth tag projection
short aPT = Short.MIN_VALUE;
double c_aPTW_hPTd = Double.NaN;
double c_aPT_hPTd = Double.NaN;
double c_hPTd = Double.NaN;
double c_aPTW_aPT = Double.NaN;
double c_aPT = Double.NaN;
if (useSmoothTagProjection) {
aPT = tagProject(dependency.arg.tag);
short hPT = tagProject(dependency.head.tag);
IntTaggedWord projectedArg = new IntTaggedWord(dependency.arg.word, aPT);
IntTaggedWord projectedAnyHead = new IntTaggedWord(ANY_WORD_INT, hPT);
IntTaggedWord projectedAnyArg = new IntTaggedWord(ANY_WORD_INT, aPT);
temp = new IntDependency(projectedAnyHead, projectedArg, leftHeaded, binDistance);
c_aPTW_hPTd = argCounter.getCount(temp);
temp = new IntDependency(projectedAnyHead, projectedAnyArg, leftHeaded, binDistance);
c_aPT_hPTd = argCounter.getCount(temp);
temp = new IntDependency(projectedAnyHead, wildTW, leftHeaded, binDistance);
c_hPTd = argCounter.getCount(temp);
temp = new IntDependency(wildTW, projectedArg, false, ANY_DISTANCE_INT);
c_aPTW_aPT = argCounter.getCount(temp);
temp = new IntDependency(wildTW, projectedAnyArg, false, ANY_DISTANCE_INT);
c_aPT = argCounter.getCount(temp);
}
// wild head is always directionless and no use distance
temp = new IntDependency(wildTW, dependency.arg, false, ANY_DISTANCE_INT);
double c_aTW = argCounter.getCount(temp);
temp = new IntDependency(wildTW, anyArg, false, ANY_DISTANCE_INT);
double c_aT = argCounter.getCount(temp);
temp = new IntDependency(wildTW, anyTagArg, false, ANY_DISTANCE_INT);
double c_aW = argCounter.getCount(temp);
// do the Bayesian magic
// MLE probs
double p_aTW_hTd;
double p_aT_hTd;
double p_aTW_aT;
double p_aW;
double p_aPTW_aPT;
double p_aPTW_hPTd;
double p_aPT_hPTd;
// backoffs either mle or themselves bayesian smoothed depending on useSmoothTagProjection
if (useSmoothTagProjection) {
if (useUnigramWordSmoothing) {
p_aW = c_aW > 0.0 ? (c_aW / numWordTokens) : 1.0; // NEED this 1.0 for unknown words!!!
p_aPTW_aPT = (c_aPTW_aPT + smooth_aPTW_aPT * p_aW) / (c_aPT + smooth_aPTW_aPT);
} else {
p_aPTW_aPT =
c_aPTW_aPT > 0.0 ? (c_aPTW_aPT / c_aPT) : 1.0; // NEED this 1.0 for unknown words!!!
}
p_aTW_aT = (c_aTW + smooth_aTW_aT * p_aPTW_aPT) / (c_aT + smooth_aTW_aT);
p_aPTW_hPTd = c_hPTd > 0.0 ? (c_aPTW_hPTd / c_hPTd) : 0.0;
p_aTW_hTd = (c_aTW_hTd + smooth_aTW_hTd * p_aPTW_hPTd) / (c_hTd + smooth_aTW_hTd);
p_aPT_hPTd = c_hPTd > 0.0 ? (c_aPT_hPTd / c_hPTd) : 0.0;
p_aT_hTd = (c_aT_hTd + smooth_aT_hTd * p_aPT_hPTd) / (c_hTd + smooth_aT_hTd);
} else {
// here word generation isn't smoothed - can't get previously unseen word with tag. Ugh.
if (op.testOptions.useLexiconToScoreDependencyPwGt) {
// We don't know the position. Now -1 means average over 0 and 1.
p_aTW_aT =
dependency.leftHeaded
? Math.exp(lex.score(dependency.arg, 1, wordIndex.get(dependency.arg.word)))
: Math.exp(lex.score(dependency.arg, -1, wordIndex.get(dependency.arg.word)));
// double oldScore = c_aTW > 0.0 ? (c_aTW / c_aT) : 1.0;
// if (oldScore == 1.0) {
// System.err.println("#### arg=" + dependency.arg + " score=" + p_aTW_aT +
// " oldScore=" + oldScore + " c_aTW=" + c_aTW + " c_aW=" + c_aW);
// }
} else {
p_aTW_aT = c_aTW > 0.0 ? (c_aTW / c_aT) : 1.0;
}
p_aTW_hTd = c_hTd > 0.0 ? (c_aTW_hTd / c_hTd) : 0.0;
p_aT_hTd = c_hTd > 0.0 ? (c_aT_hTd / c_hTd) : 0.0;
}
double pb_aTW_hTWd = (c_aTW_hTWd + smooth_aTW_hTWd * p_aTW_hTd) / (c_hTWd + smooth_aTW_hTWd);
double pb_aT_hTWd = (c_aT_hTWd + smooth_aT_hTWd * p_aT_hTd) / (c_hTWd + smooth_aT_hTWd);
double score = (interp * pb_aTW_hTWd + (1.0 - interp) * p_aTW_aT * pb_aT_hTWd) * pb_go_hTWds;
if (verbose) {
NumberFormat nf = NumberFormat.getNumberInstance();
nf.setMaximumFractionDigits(2);
if (useSmoothTagProjection) {
if (useUnigramWordSmoothing) {
System.err.println(
" c_aW=" + c_aW + ", numWordTokens=" + numWordTokens + ", p(aW)=" + nf.format(p_aW));
}
System.err.println(
" c_aPTW_aPT="
+ c_aPTW_aPT
+ ", c_aPT="
+ c_aPT
+ ", smooth_aPTW_aPT="
+ smooth_aPTW_aPT
+ ", p(aPTW|aPT)="
+ nf.format(p_aPTW_aPT));
}
System.err.println(
" c_aTW="
+ c_aTW
+ ", c_aT="
+ c_aT
+ ", smooth_aTW_aT="
+ smooth_aTW_aT
+ ", ## p(aTW|aT)="
+ nf.format(p_aTW_aT));
if (useSmoothTagProjection) {
System.err.println(
" c_aPTW_hPTd="
+ c_aPTW_hPTd
+ ", c_hPTd="
+ c_hPTd
+ ", p(aPTW|hPTd)="
+ nf.format(p_aPTW_hPTd));
}
System.err.println(
" c_aTW_hTd="
+ c_aTW_hTd
+ ", c_hTd="
+ c_hTd
+ ", smooth_aTW_hTd="
+ smooth_aTW_hTd
+ ", p(aTW|hTd)="
+ nf.format(p_aTW_hTd));
if (useSmoothTagProjection) {
System.err.println(
" c_aPT_hPTd="
+ c_aPT_hPTd
+ ", c_hPTd="
+ c_hPTd
+ ", p(aPT|hPTd)="
+ nf.format(p_aPT_hPTd));
}
System.err.println(
" c_aT_hTd="
+ c_aT_hTd
+ ", c_hTd="
+ c_hTd
+ ", smooth_aT_hTd="
+ smooth_aT_hTd
+ ", p(aT|hTd)="
+ nf.format(p_aT_hTd));
System.err.println(
" c_aTW_hTWd="
+ c_aTW_hTWd
+ ", c_hTWd="
+ c_hTWd
+ ", smooth_aTW_hTWd="
+ smooth_aTW_hTWd
+ ", ## p(aTW|hTWd)="
+ nf.format(pb_aTW_hTWd));
System.err.println(
" c_aT_hTWd="
+ c_aT_hTWd
+ ", c_hTWd="
+ c_hTWd
+ ", smooth_aT_hTWd="
+ smooth_aT_hTWd
+ ", ## p(aT|hTWd)="
+ nf.format(pb_aT_hTWd));
System.err.println(
" interp="
+ interp
+ ", prescore="
+ nf.format(interp * pb_aTW_hTWd + (1.0 - interp) * p_aTW_aT * pb_aT_hTWd)
+ ", P(go|hTWds)="
+ nf.format(pb_go_hTWds)
+ ", score="
+ nf.format(score));
}
if (op.testOptions.prunePunc && pruneTW(aTW)) {
return 1.0;
}
if (Double.isNaN(score)) {
score = 0.0;
}
// if (op.testOptions.rightBonus && ! dependency.leftHeaded)
// score -= 0.2;
if (score < MIN_PROBABILITY) {
score = 0.0;
}
return score;
}
public static Vector<Double> trainNode(
String filename,
int outputRepresentation,
String inputRepresentation,
int filter,
double learningRate,
int epochs) {
Vector<Double> node = null; // initialize node to null
double error = 1; // initialize error
double output = 0; // initialize output
Vector<Double> trainError = new Vector<Double>();
double meansq = 0;
String binTarget = "";
String binaryRep = "";
int target;
openFile(filename); // open our file for training
firstRead = true;
int targetCounter = 0;
while ((target = readFileIntoInputVector(inputRepresentation))
!= -2) { // while we have not reached out end of file error
if (outputRepresentation == 4) {
binTarget = Integer.toBinaryString(target); // convert target to binary
int[] targetArray = new int[4]; // array to hold binary ints
for (int h = 0; h < targetArray.length; h++) { // add binary elements to int array
try {
targetArray[targetArray.length - h - 1] =
Character.digit(binTarget.charAt(binTarget.length() - h - 1), 10);
} catch (IndexOutOfBoundsException e) { // if there is empty spaces, fill with zeros
targetArray[targetArray.length - h - 1] = 0;
}
}
target = targetArray[filter];
binTarget = "";
for (int i = 0; i < targetArray.length; i++) {
binTarget = binTarget + targetArray[i];
}
}
if ((filter == target && outputRepresentation == 10) || outputRepresentation != 10) {
if (node == null) {
node = new Vector<Double>();
for (int i = 0; i < inputs.size(); i++) { // for each input
double randWeight = 1; // initialize random weight
while (randWeight > 0.15) // while out random isn't less than 0.15
randWeight = random.nextDouble(); // get a new random double
if (random.nextBoolean()) randWeight = randWeight * -1; // randomly set to negative
node.addElement(new Double(randWeight)); // store random weight
}
}
targetCounter++;
// train for number of epochs
for (int i = 0; i < epochs; i++) { // for each epoch
double weightedSum = 0; // set/reset the weighted sum to 0
for (int j = 0; j < inputs.size(); j++) {
weightedSum += (inputs.elementAt(j) * node.elementAt(j)); // calculate weighted sum
}
output = activation(weightedSum); // retrieve output
if (outputRepresentation == 1 || outputRepresentation == 10) output = output * 10;
else if (outputRepresentation == 4) { // each node must have a whole number
if (output > 0.5) output = 1;
else output = 0;
}
error = target - output; // calculate error
if (i == epochs - 1) {
if (outputRepresentation == 4) {
if (filter == 0) {
allOutputs.ensureCapacity(targetCounter);
allOutputs.add(Integer.toString((int) output));
} else {
allOutputs.set(
targetCounter - 1, allOutputs.get(targetCounter - 1) + ((int) output));
}
}
trainError.add(error);
if (Math.round(error) == 0 && outputRepresentation != 4) {
trainCountCorrect++;
} else if (outputRepresentation == 4 && filter == 3) {
if (allOutputs.get(targetCounter - 1).equals(binTarget)) {
trainCountCorrect++;
}
}
}
for (int k = 0; k < inputs.size(); k++) { // for each input
double derivative =
(1.64872 * Math.exp(weightedSum))
/ (Math.pow(1.64872 + Math.exp(weightedSum), 2)); // calculate new weights
double newWeight =
node.elementAt(k)
+ (learningRate * inputs.elementAt(k) * error * derivative); // cont.
node.setElementAt(newWeight, k); // update weights
}
}
}
inputs.clear();
}
errorVect.addElement((Vector<Double>) trainError.clone());
trainError.clear();
if (outputRepresentation == 4 && filter == 3)
System.out.println(
"Training Percent: " + (100 * (double) trainCountCorrect / (double) targetCounter));
if (outputRepresentation == 10 && filter == 9)
System.out.println(
"Training Percent: " + (10 * (double) trainCountCorrect / (double) targetCounter));
if (outputRepresentation == 1)
System.out.println(
"Training Percent: " + (100 * (double) trainCountCorrect / (double) targetCounter));
closeFile();
return node; // node is now trained for an epoch
}