rtrees.cpp 14.6 KB
Newer Older
wester committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424
/*M///////////////////////////////////////////////////////////////////////////////////////
//
//  IMPORTANT: READ BEFORE DOWNLOADING, COPYING, INSTALLING OR USING.
//
//  By downloading, copying, installing or using the software you agree to this license.
//  If you do not agree to this license, do not download, install,
//  copy or use the software.
//
//
//                           License Agreement
//                For Open Source Computer Vision Library
//
// Copyright (C) 2000, Intel Corporation, all rights reserved.
// Copyright (C) 2014, Itseez Inc, all rights reserved.
// Third party copyrights are property of their respective owners.
//
// Redistribution and use in source and binary forms, with or without modification,
// are permitted provided that the following conditions are met:
//
//   * Redistribution's of source code must retain the above copyright notice,
//     this list of conditions and the following disclaimer.
//
//   * Redistribution's in binary form must reproduce the above copyright notice,
//     this list of conditions and the following disclaimer in the documentation
//     and/or other materials provided with the distribution.
//
//   * The name of the copyright holders may not be used to endorse or promote products
//     derived from this software without specific prior written permission.
//
// This software is provided by the copyright holders and contributors "as is" and
// any express or implied warranties, including, but not limited to, the implied
// warranties of merchantability and fitness for a particular purpose are disclaimed.
// In no event shall the Intel Corporation or contributors be liable for any direct,
// indirect, incidental, special, exemplary, or consequential damages
// (including, but not limited to, procurement of substitute goods or services;
// loss of use, data, or profits; or business interruption) however caused
// and on any theory of liability, whether in contract, strict liability,
// or tort (including negligence or otherwise) arising in any way out of
// the use of this software, even if advised of the possibility of such damage.
//
//M*/

#include "precomp.hpp"

namespace cv {
namespace ml {

//////////////////////////////////////////////////////////////////////////////////////////
//                                  Random trees                                        //
//////////////////////////////////////////////////////////////////////////////////////////
RTreeParams::RTreeParams()
{
    calcVarImportance = false;
    nactiveVars = 0;
    termCrit = TermCriteria(TermCriteria::EPS + TermCriteria::COUNT, 50, 0.1);
}

RTreeParams::RTreeParams(bool _calcVarImportance,
                         int _nactiveVars,
                         TermCriteria _termCrit )
{
    calcVarImportance = _calcVarImportance;
    nactiveVars = _nactiveVars;
    termCrit = _termCrit;
}


class DTreesImplForRTrees : public DTreesImpl
{
public:
    DTreesImplForRTrees()
    {
        params.setMaxDepth(5);
        params.setMinSampleCount(10);
        params.setRegressionAccuracy(0.f);
        params.useSurrogates = false;
        params.setMaxCategories(10);
        params.setCVFolds(0);
        params.use1SERule = false;
        params.truncatePrunedTree = false;
        params.priors = Mat();
    }
    virtual ~DTreesImplForRTrees() {}

    void clear()
    {
        DTreesImpl::clear();
        oobError = 0.;
        rng = RNG((uint64)-1);
    }

    const vector<int>& getActiveVars()
    {
        int i, nvars = (int)allVars.size(), m = (int)activeVars.size();
        for( i = 0; i < nvars; i++ )
        {
            int i1 = rng.uniform(0, nvars);
            int i2 = rng.uniform(0, nvars);
            std::swap(allVars[i1], allVars[i2]);
        }
        for( i = 0; i < m; i++ )
            activeVars[i] = allVars[i];
        return activeVars;
    }

    void startTraining( const Ptr<TrainData>& trainData, int flags )
    {
        DTreesImpl::startTraining(trainData, flags);
        int nvars = w->data->getNVars();
        int i, m = rparams.nactiveVars > 0 ? rparams.nactiveVars : cvRound(std::sqrt((double)nvars));
        m = std::min(std::max(m, 1), nvars);
        allVars.resize(nvars);
        activeVars.resize(m);
        for( i = 0; i < nvars; i++ )
            allVars[i] = varIdx[i];
    }

    void endTraining()
    {
        DTreesImpl::endTraining();
        vector<int> a, b;
        std::swap(allVars, a);
        std::swap(activeVars, b);
    }

    bool train( const Ptr<TrainData>& trainData, int flags )
    {
        startTraining(trainData, flags);
        int treeidx, ntrees = (rparams.termCrit.type & TermCriteria::COUNT) != 0 ?
            rparams.termCrit.maxCount : 10000;
        int i, j, k, vi, vi_, n = (int)w->sidx.size();
        int nclasses = (int)classLabels.size();
        double eps = (rparams.termCrit.type & TermCriteria::EPS) != 0 &&
            rparams.termCrit.epsilon > 0 ? rparams.termCrit.epsilon : 0.;
        vector<int> sidx(n);
        vector<uchar> oobmask(n);
        vector<int> oobidx;
        vector<int> oobperm;
        vector<double> oobres(n, 0.);
        vector<int> oobcount(n, 0);
        vector<int> oobvotes(n*nclasses, 0);
        int nvars = w->data->getNVars();
        int nallvars = w->data->getNAllVars();
        const int* vidx = !varIdx.empty() ? &varIdx[0] : 0;
        vector<float> samplebuf(nallvars);
        Mat samples = w->data->getSamples();
        float* psamples = samples.ptr<float>();
        size_t sstep0 = samples.step1(), sstep1 = 1;
        Mat sample0, sample(nallvars, 1, CV_32F, &samplebuf[0]);
        int predictFlags = _isClassifier ? (PREDICT_MAX_VOTE + RAW_OUTPUT) : PREDICT_SUM;

        bool calcOOBError = eps > 0 || rparams.calcVarImportance;
        double max_response = 0.;

        if( w->data->getLayout() == COL_SAMPLE )
            std::swap(sstep0, sstep1);

        if( !_isClassifier )
        {
            for( i = 0; i < n; i++ )
            {
                double val = std::abs(w->ord_responses[w->sidx[i]]);
                max_response = std::max(max_response, val);
            }
        }

        if( rparams.calcVarImportance )
            varImportance.resize(nallvars, 0.f);

        for( treeidx = 0; treeidx < ntrees; treeidx++ )
        {
            for( i = 0; i < n; i++ )
                oobmask[i] = (uchar)1;

            for( i = 0; i < n; i++ )
            {
                j = rng.uniform(0, n);
                sidx[i] = w->sidx[j];
                oobmask[j] = (uchar)0;
            }
            int root = addTree( sidx );
            if( root < 0 )
                return false;

            if( calcOOBError )
            {
                oobidx.clear();
                for( i = 0; i < n; i++ )
                {
                    if( oobmask[i] )
                        oobidx.push_back(i);
                }
                int n_oob = (int)oobidx.size();
                // if there is no out-of-bag samples, we can not compute OOB error
                // nor update the variable importance vector; so we proceed to the next tree
                if( n_oob == 0 )
                    continue;
                double ncorrect_responses = 0.;

                oobError = 0.;
                for( i = 0; i < n_oob; i++ )
                {
                    j = oobidx[i];
                    sample = Mat( nallvars, 1, CV_32F, psamples + sstep0*w->sidx[j], sstep1*sizeof(psamples[0]) );

                    double val = predictTrees(Range(treeidx, treeidx+1), sample, predictFlags);
                    if( !_isClassifier )
                    {
                        oobres[j] += val;
                        oobcount[j]++;
                        double true_val = w->ord_responses[w->sidx[j]];
                        double a = oobres[j]/oobcount[j] - true_val;
                        oobError += a*a;
                        val = (val - true_val)/max_response;
                        ncorrect_responses += std::exp( -val*val );
                    }
                    else
                    {
                        int ival = cvRound(val);
                        //Voting scheme to combine OOB errors of each tree
                        int* votes = &oobvotes[j*nclasses];
                        votes[ival]++;
                        int best_class = 0;
                        for( k = 1; k < nclasses; k++ )
                            if( votes[best_class] < votes[k] )
                                best_class = k;
                        int diff = best_class != w->cat_responses[w->sidx[j]];
                        oobError += diff;
                        ncorrect_responses += diff == 0;
                    }
                }

                oobError /= n_oob;
                if( rparams.calcVarImportance && n_oob > 1 )
                {
                    Mat sample_clone;
                    oobperm.resize(n_oob);
                    for( i = 0; i < n_oob; i++ )
                        oobperm[i] = oobidx[i];
                    for (i = n_oob - 1; i > 0; --i)  //Randomly shuffle indices so we can permute features
                    {
                        int r_i = rng.uniform(0, n_oob);
                        std::swap(oobperm[i], oobperm[r_i]);
                    }

                    for( vi_ = 0; vi_ < nvars; vi_++ )
                    {
                        vi = vidx ? vidx[vi_] : vi_; //Ensure that only the user specified predictors are used for training
                        double ncorrect_responses_permuted = 0;

                        for( i = 0; i < n_oob; i++ )
                        {
                            j = oobidx[i];
                            int vj = oobperm[i];
                            sample0 = Mat( nallvars, 1, CV_32F, psamples + sstep0*w->sidx[j], sstep1*sizeof(psamples[0]) );
                            sample0.copyTo(sample_clone); //create a copy so we don't mess up the original data
                            sample_clone.at<float>(vi) = psamples[sstep0*w->sidx[vj] + sstep1*vi];

                            double val = predictTrees(Range(treeidx, treeidx+1), sample_clone, predictFlags);
                            if( !_isClassifier )
                            {
                                val = (val - w->ord_responses[w->sidx[j]])/max_response;
                                ncorrect_responses_permuted += exp( -val*val );
                            }
                            else
                            {
                                ncorrect_responses_permuted += cvRound(val) == w->cat_responses[w->sidx[j]];
                            }
                        }
                        varImportance[vi] += (float)(ncorrect_responses - ncorrect_responses_permuted);
                    }
                }
            }
            if( calcOOBError && oobError < eps )
                break;
        }

        if( rparams.calcVarImportance )
        {
            for( vi_ = 0; vi_ < nallvars; vi_++ )
                varImportance[vi_] = std::max(varImportance[vi_], 0.f);
            normalize(varImportance, varImportance, 1., 0, NORM_L1);
        }
        endTraining();
        return true;
    }

    void writeTrainingParams( FileStorage& fs ) const
    {
        DTreesImpl::writeTrainingParams(fs);
        fs << "nactive_vars" << rparams.nactiveVars;
    }

    void write( FileStorage& fs ) const
    {
        if( roots.empty() )
            CV_Error( CV_StsBadArg, "RTrees have not been trained" );

        writeParams(fs);

        fs << "oob_error" << oobError;
        if( !varImportance.empty() )
            fs << "var_importance" << varImportance;

        int k, ntrees = (int)roots.size();

        fs << "ntrees" << ntrees
           << "trees" << "[";

        for( k = 0; k < ntrees; k++ )
        {
            fs << "{";
            writeTree(fs, roots[k]);
            fs << "}";
        }

        fs << "]";
    }

    void readParams( const FileNode& fn )
    {
        DTreesImpl::readParams(fn);

        FileNode tparams_node = fn["training_params"];
        rparams.nactiveVars = (int)tparams_node["nactive_vars"];
    }

    void read( const FileNode& fn )
    {
        clear();

        //int nclasses = (int)fn["nclasses"];
        //int nsamples = (int)fn["nsamples"];
        oobError = (double)fn["oob_error"];
        int ntrees = (int)fn["ntrees"];

        readVectorOrMat(fn["var_importance"], varImportance);

        readParams(fn);

        FileNode trees_node = fn["trees"];
        FileNodeIterator it = trees_node.begin();
        CV_Assert( ntrees == (int)trees_node.size() );

        for( int treeidx = 0; treeidx < ntrees; treeidx++, ++it )
        {
            FileNode nfn = (*it)["nodes"];
            readTree(nfn);
        }
    }

    RTreeParams rparams;
    double oobError;
    vector<float> varImportance;
    vector<int> allVars, activeVars;
    RNG rng;
};


class RTreesImpl : public RTrees
{
public:
    CV_IMPL_PROPERTY(bool, CalculateVarImportance, impl.rparams.calcVarImportance)
    CV_IMPL_PROPERTY(int, ActiveVarCount, impl.rparams.nactiveVars)
    CV_IMPL_PROPERTY_S(TermCriteria, TermCriteria, impl.rparams.termCrit)

    CV_WRAP_SAME_PROPERTY(int, MaxCategories, impl.params)
    CV_WRAP_SAME_PROPERTY(int, MaxDepth, impl.params)
    CV_WRAP_SAME_PROPERTY(int, MinSampleCount, impl.params)
    CV_WRAP_SAME_PROPERTY(int, CVFolds, impl.params)
    CV_WRAP_SAME_PROPERTY(bool, UseSurrogates, impl.params)
    CV_WRAP_SAME_PROPERTY(bool, Use1SERule, impl.params)
    CV_WRAP_SAME_PROPERTY(bool, TruncatePrunedTree, impl.params)
    CV_WRAP_SAME_PROPERTY(float, RegressionAccuracy, impl.params)
    CV_WRAP_SAME_PROPERTY_S(cv::Mat, Priors, impl.params)

    RTreesImpl() {}
    virtual ~RTreesImpl() {}

    String getDefaultName() const { return "opencv_ml_rtrees"; }

    bool train( const Ptr<TrainData>& trainData, int flags )
    {
        return impl.train(trainData, flags);
    }

    float predict( InputArray samples, OutputArray results, int flags ) const
    {
        return impl.predict(samples, results, flags);
    }

    void write( FileStorage& fs ) const
    {
        impl.write(fs);
    }

    void read( const FileNode& fn )
    {
        impl.read(fn);
    }

    Mat getVarImportance() const { return Mat_<float>(impl.varImportance, true); }
    int getVarCount() const { return impl.getVarCount(); }

    bool isTrained() const { return impl.isTrained(); }
    bool isClassifier() const { return impl.isClassifier(); }

    const vector<int>& getRoots() const { return impl.getRoots(); }
    const vector<Node>& getNodes() const { return impl.getNodes(); }
    const vector<Split>& getSplits() const { return impl.getSplits(); }
    const vector<int>& getSubsets() const { return impl.getSubsets(); }

    DTreesImplForRTrees impl;
};


Ptr<RTrees> RTrees::create()
{
    return makePtr<RTreesImpl>();
}

}}

// End of file.