/*
* Hivemall: Hive scalable Machine Learning Library
*
* Copyright (C) 2013
* National Institute of Advanced Industrial Science and Technology (AIST)
* Registration Number: H25PRO-1520
*
* 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.
*
* 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 Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
package hivemall.classifier.multiclass;
import hivemall.io.FeatureValue;
import hivemall.io.IWeightValue;
import hivemall.io.Margin;
import hivemall.io.PredictionModel;
import hivemall.io.WeightValue.WeightValueWithCovar;
import hivemall.utils.math.StatsUtils;
import java.util.List;
import org.apache.commons.cli.CommandLine;
import org.apache.commons.cli.Options;
import org.apache.hadoop.hive.ql.exec.UDFArgumentException;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspectorUtils;
import org.apache.hadoop.hive.serde2.objectinspector.StructObjectInspector;
/**
* Soft Confidence-Weighted binary classifier.
* <pre>
* [1] Steven C. H. Hoi, Jialei Wang, Peilin Zhao: Exact Soft Confidence-Weighted Learning. ICML 2012
* </pre>
*
* @link http://icml.cc/2012/papers/86.pdf
*/
public abstract class MulticlassSoftConfidenceWeightedUDTF extends MulticlassOnlineClassifierUDTF {
/** Confidence parameter phi */
protected float phi;
/** Aggressiveness parameter */
protected float c;
@Override
public StructObjectInspector initialize(ObjectInspector[] argOIs) throws UDFArgumentException {
final int numArgs = argOIs.length;
if(numArgs != 2 && numArgs != 3) {
throw new UDFArgumentException("MulticlassSoftConfidenceWeightedUDTF takes 2 or 3 arguments: List<String|Int|BitInt> features, {Int|String} label [, constant String options]");
}
return super.initialize(argOIs);
}
@Override
protected boolean useCovariance() {
return true;
}
@Override
protected Options getOptions() {
Options opts = super.getOptions();
opts.addOption("phi", "confidence", true, "Confidence parameter [default 1.0]");
opts.addOption("eta", "hyper_c", true, "Confidence hyperparameter eta in range (0.5, 1] [default 0.85]");
opts.addOption("c", "aggressiveness", true, "Aggressiveness parameter C [default 1.0]");
return opts;
}
@Override
protected CommandLine processOptions(ObjectInspector[] argOIs) throws UDFArgumentException {
final CommandLine cl = super.processOptions(argOIs);
float phi = 1.f;
float c = 1.f;
if(cl != null) {
String phi_str = cl.getOptionValue("phi");
if(phi_str == null) {
String eta_str = cl.getOptionValue("eta");
if(eta_str != null) {
double eta = Double.parseDouble(eta_str);
if(eta <= 0.5 || eta > 1) {
throw new UDFArgumentException("Confidence hyperparameter eta must be in range (0.5, 1]: "
+ eta_str);
}
phi = (float) StatsUtils.probit(eta, 5d);
}
} else {
phi = Float.parseFloat(phi_str);
}
String c_str = cl.getOptionValue("c");
if(c_str != null) {
c = Float.parseFloat(c_str);
if(!(c > 0.f)) {
throw new UDFArgumentException("Aggressiveness parameter C must be C > 0: " + c);
}
}
}
this.phi = phi;
this.c = c;
return cl;
}
@Override
protected void train(List<?> features, Object actual_label) {
Margin margin = getMarginAndVariance(features, actual_label, true);
float loss = loss(margin);
if(loss > 0.f) {
float alpha = getAlpha(margin);
if(alpha == 0.f) {
return;
}
float beta = getBeta(margin, alpha);
if(beta == 0.f) {
return;
}
Object missed_label = margin.getMaxIncorrectLabel();
update(features, actual_label, missed_label, alpha, beta);
}
}
protected float loss(Margin margin) {
float var = margin.getVariance();
float m = margin.get();
assert (var != 0);
float loss = phi * (float) Math.sqrt(var) - m;
return Math.max(loss, 0.f);
}
protected abstract float getAlpha(Margin margin);
protected abstract float getBeta(Margin margin, float alpha);
public static class SCW1 extends MulticlassSoftConfidenceWeightedUDTF {
private float squared_phi, psi, zeta;
@Override
public StructObjectInspector initialize(ObjectInspector[] argOIs)
throws UDFArgumentException {
StructObjectInspector oi = super.initialize(argOIs);
float phiphi = phi * phi;
this.squared_phi = phiphi;
this.psi = 1.f + phiphi / 2.f;
this.zeta = 1.f + phiphi;
return oi;
}
@Override
protected float getAlpha(Margin margin) {
float m = margin.get();
float var = margin.getVariance();
float alpha_numer = -m
* psi
+ (float) Math.sqrt((m * m * squared_phi * squared_phi / 4.f)
+ (var * squared_phi * zeta));
float alpha_denom = var * zeta;
if(alpha_denom == 0.f) {
return 0.f;
}
float alpha = alpha_numer / alpha_denom;
if(alpha <= 0.f) {
return 0.f;
}
return Math.max(c, alpha);
}
@Override
protected float getBeta(Margin margin, float alpha) {
if(alpha == 0.f) {
return 0.f;
}
float var = margin.getVariance();
float beta_numer = alpha * phi;
float var_alpha_phi = var * beta_numer;
float u = -var_alpha_phi + (float) Math.sqrt(var_alpha_phi * var_alpha_phi + 4.f * var);
float beta_den = u / 2.f + var_alpha_phi;
if(beta_den == 0.f) {
return 0.f;
}
float beta = beta_numer / beta_den;
return beta;
}
}
public static class SCW2 extends SCW1 {
@Override
protected float getAlpha(Margin margin) {
float m = margin.get();
float var = margin.getVariance();
float squared_phi = phi * phi;
float n = var + c / 2.f;
float v_phi_phi = var * squared_phi;
float v_phi_phi_m = v_phi_phi * m;
float term = v_phi_phi_m * m * var + 4.f * n * var * (n + v_phi_phi);
float gamma = phi * (float) Math.sqrt(term);
float alpha_numer = -(2.f * m * n + v_phi_phi_m) + gamma;
if(alpha_numer <= 0.f) {
return 0.f;
}
float alpha_denom = 2.f * (n * n + n * v_phi_phi);
if(alpha_denom == 0.f) {
return 0.f;
}
float alpha = alpha_numer / alpha_denom;
return Math.max(0.f, alpha);
}
}
protected void update(final List<?> features, final Object actual_label, final Object missed_label, final float alpha, final float beta) {
assert (actual_label != null);
if(actual_label.equals(missed_label)) {
throw new IllegalArgumentException("Actual label equals to missed label: "
+ actual_label);
}
PredictionModel model2add = label2model.get(actual_label);
if(model2add == null) {
model2add = createModel();
label2model.put(actual_label, model2add);
}
PredictionModel model2sub = null;
if(missed_label != null) {
model2sub = label2model.get(missed_label);
if(model2sub == null) {
model2sub = createModel();
label2model.put(missed_label, model2sub);
}
}
final ObjectInspector featureInspector = featureListOI.getListElementObjectInspector();
for(Object f : features) {// w[f] += y * x[f]
if(f == null) {
continue;
}
final Object k;
final float v;
if(parseFeature) {
FeatureValue fv = FeatureValue.parse(f);
k = fv.getFeature();
v = fv.getValue();
} else {
k = ObjectInspectorUtils.copyToStandardObject(f, featureInspector);
v = 1.f;
}
IWeightValue old_correctclass_w = model2add.get(k);
IWeightValue new_correctclass_w = getNewWeight(old_correctclass_w, v, alpha, beta, true);
model2add.set(k, new_correctclass_w);
if(model2sub != null) {
IWeightValue old_wrongclass_w = model2sub.get(k);
IWeightValue new_wrongclass_w = getNewWeight(old_wrongclass_w, v, alpha, beta, false);
model2sub.set(k, new_wrongclass_w);
}
}
}
private static IWeightValue getNewWeight(final IWeightValue old, final float v, final float alpha, final float beta, final boolean positive) {
final float old_v;
final float old_cov;
if(old == null) {
old_v = 0.f;
old_cov = 1.f;
} else {
old_v = old.get();
old_cov = old.getCovariance();
}
float cv = old_cov * v;
float new_w = positive ? old_v + (alpha * cv) : old_v - (alpha * cv);
float new_cov = old_cov - (beta * cv * cv);
return new WeightValueWithCovar(new_w, new_cov);
}
}