• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 // Ceres Solver - A fast non-linear least squares minimizer
2 // Copyright 2012 Google Inc. All rights reserved.
3 // http://code.google.com/p/ceres-solver/
4 //
5 // Redistribution and use in source and binary forms, with or without
6 // modification, are permitted provided that the following conditions are met:
7 //
8 // * Redistributions of source code must retain the above copyright notice,
9 //   this list of conditions and the following disclaimer.
10 // * Redistributions in binary form must reproduce the above copyright notice,
11 //   this list of conditions and the following disclaimer in the documentation
12 //   and/or other materials provided with the distribution.
13 // * Neither the name of Google Inc. nor the names of its contributors may be
14 //   used to endorse or promote products derived from this software without
15 //   specific prior written permission.
16 //
17 // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 // AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 // IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 // ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21 // LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 // CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 // SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 // INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 // CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 // ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 // POSSIBILITY OF SUCH DAMAGE.
28 //
29 // Author: sameeragarwal@google.com (Sameer Agarwal)
30 //
31 // The National Institute of Standards and Technology has released a
32 // set of problems to test non-linear least squares solvers.
33 //
34 // More information about the background on these problems and
35 // suggested evaluation methodology can be found at:
36 //
37 //   http://www.itl.nist.gov/div898/strd/nls/nls_info.shtml
38 //
39 // The problem data themselves can be found at
40 //
41 //   http://www.itl.nist.gov/div898/strd/nls/nls_main.shtml
42 //
43 // The problems are divided into three levels of difficulty, Easy,
44 // Medium and Hard. For each problem there are two starting guesses,
45 // the first one far away from the global minimum and the second
46 // closer to it.
47 //
48 // A problem is considered successfully solved, if every components of
49 // the solution matches the globally optimal solution in at least 4
50 // digits or more.
51 //
52 // This dataset was used for an evaluation of Non-linear least squares
53 // solvers:
54 //
55 // P. F. Mondragon & B. Borchers, A Comparison of Nonlinear Regression
56 // Codes, Journal of Modern Applied Statistical Methods, 4(1):343-351,
57 // 2005.
58 //
59 // The results from Mondragon & Borchers can be summarized as
60 //               Excel  Gnuplot  GaussFit  HBN  MinPack
61 // Average LRE     2.3      4.3       4.0  6.8      4.4
62 //      Winner       1        5        12   29       12
63 //
64 // Where the row Winner counts, the number of problems for which the
65 // solver had the highest LRE.
66 
67 // In this file, we implement the same evaluation methodology using
68 // Ceres. Currently using Levenberg-Marquard with DENSE_QR, we get
69 //
70 //               Excel  Gnuplot  GaussFit  HBN  MinPack  Ceres
71 // Average LRE     2.3      4.3       4.0  6.8      4.4    9.4
72 //      Winner       0        0         5   11        2     41
73 
74 #include <iostream>
75 #include <iterator>
76 #include <fstream>
77 #include "ceres/ceres.h"
78 #include "gflags/gflags.h"
79 #include "glog/logging.h"
80 #include "Eigen/Core"
81 
82 DEFINE_string(nist_data_dir, "", "Directory containing the NIST non-linear"
83               "regression examples");
84 DEFINE_string(minimizer, "trust_region",
85               "Minimizer type to use, choices are: line_search & trust_region");
86 DEFINE_string(trust_region_strategy, "levenberg_marquardt",
87               "Options are: levenberg_marquardt, dogleg");
88 DEFINE_string(dogleg, "traditional_dogleg",
89               "Options are: traditional_dogleg, subspace_dogleg");
90 DEFINE_string(linear_solver, "dense_qr", "Options are: "
91               "sparse_cholesky, dense_qr, dense_normal_cholesky and"
92               "cgnr");
93 DEFINE_string(preconditioner, "jacobi", "Options are: "
94               "identity, jacobi");
95 DEFINE_string(line_search, "armijo",
96               "Line search algorithm to use, choices are: armijo and wolfe.");
97 DEFINE_string(line_search_direction, "lbfgs",
98               "Line search direction algorithm to use, choices: lbfgs, bfgs");
99 DEFINE_int32(max_line_search_iterations, 20,
100              "Maximum number of iterations for each line search.");
101 DEFINE_int32(max_line_search_restarts, 10,
102              "Maximum number of restarts of line search direction algorithm.");
103 DEFINE_string(line_search_interpolation, "cubic",
104               "Degree of polynomial aproximation in line search, "
105               "choices are: bisection, quadratic & cubic.");
106 DEFINE_int32(lbfgs_rank, 20,
107              "Rank of L-BFGS inverse Hessian approximation in line search.");
108 DEFINE_bool(approximate_eigenvalue_bfgs_scaling, false,
109             "Use approximate eigenvalue scaling in (L)BFGS line search.");
110 DEFINE_double(sufficient_decrease, 1.0e-4,
111               "Line search Armijo sufficient (function) decrease factor.");
112 DEFINE_double(sufficient_curvature_decrease, 0.9,
113               "Line search Wolfe sufficient curvature decrease factor.");
114 DEFINE_int32(num_iterations, 10000, "Number of iterations");
115 DEFINE_bool(nonmonotonic_steps, false, "Trust region algorithm can use"
116             " nonmonotic steps");
117 DEFINE_double(initial_trust_region_radius, 1e4, "Initial trust region radius");
118 
119 namespace ceres {
120 namespace examples {
121 
122 using Eigen::Dynamic;
123 using Eigen::RowMajor;
124 typedef Eigen::Matrix<double, Dynamic, 1> Vector;
125 typedef Eigen::Matrix<double, Dynamic, Dynamic, RowMajor> Matrix;
126 
SplitStringUsingChar(const string & full,const char delim,vector<string> * result)127 void SplitStringUsingChar(const string& full,
128                           const char delim,
129                           vector<string>* result) {
130   back_insert_iterator< vector<string> > it(*result);
131 
132   const char* p = full.data();
133   const char* end = p + full.size();
134   while (p != end) {
135     if (*p == delim) {
136       ++p;
137     } else {
138       const char* start = p;
139       while (++p != end && *p != delim) {
140         // Skip to the next occurence of the delimiter.
141       }
142       *it++ = string(start, p - start);
143     }
144   }
145 }
146 
GetAndSplitLine(std::ifstream & ifs,std::vector<std::string> * pieces)147 bool GetAndSplitLine(std::ifstream& ifs, std::vector<std::string>* pieces) {
148   pieces->clear();
149   char buf[256];
150   ifs.getline(buf, 256);
151   SplitStringUsingChar(std::string(buf), ' ', pieces);
152   return true;
153 }
154 
SkipLines(std::ifstream & ifs,int num_lines)155 void SkipLines(std::ifstream& ifs, int num_lines) {
156   char buf[256];
157   for (int i = 0; i < num_lines; ++i) {
158     ifs.getline(buf, 256);
159   }
160 }
161 
162 class NISTProblem {
163  public:
NISTProblem(const std::string & filename)164   explicit NISTProblem(const std::string& filename) {
165     std::ifstream ifs(filename.c_str(), std::ifstream::in);
166 
167     std::vector<std::string> pieces;
168     SkipLines(ifs, 24);
169     GetAndSplitLine(ifs, &pieces);
170     const int kNumResponses = std::atoi(pieces[1].c_str());
171 
172     GetAndSplitLine(ifs, &pieces);
173     const int kNumPredictors = std::atoi(pieces[0].c_str());
174 
175     GetAndSplitLine(ifs, &pieces);
176     const int kNumObservations = std::atoi(pieces[0].c_str());
177 
178     SkipLines(ifs, 4);
179     GetAndSplitLine(ifs, &pieces);
180     const int kNumParameters = std::atoi(pieces[0].c_str());
181     SkipLines(ifs, 8);
182 
183     // Get the first line of initial and final parameter values to
184     // determine the number of tries.
185     GetAndSplitLine(ifs, &pieces);
186     const int kNumTries = pieces.size() - 4;
187 
188     predictor_.resize(kNumObservations, kNumPredictors);
189     response_.resize(kNumObservations, kNumResponses);
190     initial_parameters_.resize(kNumTries, kNumParameters);
191     final_parameters_.resize(1, kNumParameters);
192 
193     // Parse the line for parameter b1.
194     int parameter_id = 0;
195     for (int i = 0; i < kNumTries; ++i) {
196       initial_parameters_(i, parameter_id) = std::atof(pieces[i + 2].c_str());
197     }
198     final_parameters_(0, parameter_id) = std::atof(pieces[2 + kNumTries].c_str());
199 
200     // Parse the remaining parameter lines.
201     for (int parameter_id = 1; parameter_id < kNumParameters; ++parameter_id) {
202      GetAndSplitLine(ifs, &pieces);
203      // b2, b3, ....
204      for (int i = 0; i < kNumTries; ++i) {
205        initial_parameters_(i, parameter_id) = std::atof(pieces[i + 2].c_str());
206      }
207      final_parameters_(0, parameter_id) = std::atof(pieces[2 + kNumTries].c_str());
208     }
209 
210     // Certfied cost
211     SkipLines(ifs, 1);
212     GetAndSplitLine(ifs, &pieces);
213     certified_cost_ = std::atof(pieces[4].c_str()) / 2.0;
214 
215     // Read the observations.
216     SkipLines(ifs, 18 - kNumParameters);
217     for (int i = 0; i < kNumObservations; ++i) {
218       GetAndSplitLine(ifs, &pieces);
219       // Response.
220       for (int j = 0; j < kNumResponses; ++j) {
221         response_(i, j) =  std::atof(pieces[j].c_str());
222       }
223 
224       // Predictor variables.
225       for (int j = 0; j < kNumPredictors; ++j) {
226         predictor_(i, j) =  std::atof(pieces[j + kNumResponses].c_str());
227       }
228     }
229   }
230 
initial_parameters(int start) const231   Matrix initial_parameters(int start) const { return initial_parameters_.row(start); }
final_parameters() const232   Matrix final_parameters() const  { return final_parameters_; }
predictor() const233   Matrix predictor()        const { return predictor_;         }
response() const234   Matrix response()         const { return response_;          }
predictor_size() const235   int predictor_size()      const { return predictor_.cols();  }
num_observations() const236   int num_observations()    const { return predictor_.rows();  }
response_size() const237   int response_size()       const { return response_.cols();   }
num_parameters() const238   int num_parameters()      const { return initial_parameters_.cols(); }
num_starts() const239   int num_starts()          const { return initial_parameters_.rows(); }
certified_cost() const240   double certified_cost()   const { return certified_cost_; }
241 
242  private:
243   Matrix predictor_;
244   Matrix response_;
245   Matrix initial_parameters_;
246   Matrix final_parameters_;
247   double certified_cost_;
248 };
249 
250 #define NIST_BEGIN(CostFunctionName) \
251   struct CostFunctionName { \
252     CostFunctionName(const double* const x, \
253                      const double* const y) \
254         : x_(*x), y_(*y) {} \
255     double x_; \
256     double y_; \
257     template <typename T> \
258     bool operator()(const T* const b, T* residual) const { \
259     const T y(y_); \
260     const T x(x_); \
261       residual[0] = y - (
262 
263 #define NIST_END ); return true; }};
264 
265 // y = b1 * (b2+x)**(-1/b3)  +  e
266 NIST_BEGIN(Bennet5)
267   b[0] * pow(b[1] + x, T(-1.0) / b[2])
268 NIST_END
269 
270 // y = b1*(1-exp[-b2*x])  +  e
271 NIST_BEGIN(BoxBOD)
272   b[0] * (T(1.0) - exp(-b[1] * x))
273 NIST_END
274 
275 // y = exp[-b1*x]/(b2+b3*x)  +  e
276 NIST_BEGIN(Chwirut)
277   exp(-b[0] * x) / (b[1] + b[2] * x)
278 NIST_END
279 
280 // y  = b1*x**b2  +  e
281 NIST_BEGIN(DanWood)
282   b[0] * pow(x, b[1])
283 NIST_END
284 
285 // y = b1*exp( -b2*x ) + b3*exp( -(x-b4)**2 / b5**2 )
286 //     + b6*exp( -(x-b7)**2 / b8**2 ) + e
287 NIST_BEGIN(Gauss)
288   b[0] * exp(-b[1] * x) +
289   b[2] * exp(-pow((x - b[3])/b[4], 2)) +
290   b[5] * exp(-pow((x - b[6])/b[7],2))
291 NIST_END
292 
293 // y = b1*exp(-b2*x) + b3*exp(-b4*x) + b5*exp(-b6*x)  +  e
294 NIST_BEGIN(Lanczos)
295   b[0] * exp(-b[1] * x) + b[2] * exp(-b[3] * x) + b[4] * exp(-b[5] * x)
296 NIST_END
297 
298 // y = (b1+b2*x+b3*x**2+b4*x**3) /
299 //     (1+b5*x+b6*x**2+b7*x**3)  +  e
300 NIST_BEGIN(Hahn1)
301   (b[0] + b[1] * x + b[2] * x * x + b[3] * x * x * x) /
302   (T(1.0) + b[4] * x + b[5] * x * x + b[6] * x * x * x)
303 NIST_END
304 
305 // y = (b1 + b2*x + b3*x**2) /
306 //    (1 + b4*x + b5*x**2)  +  e
307 NIST_BEGIN(Kirby2)
308   (b[0] + b[1] * x + b[2] * x * x) /
309   (T(1.0) + b[3] * x + b[4] * x * x)
310 NIST_END
311 
312 // y = b1*(x**2+x*b2) / (x**2+x*b3+b4)  +  e
313 NIST_BEGIN(MGH09)
314   b[0] * (x * x + x * b[1]) / (x * x + x * b[2] + b[3])
315 NIST_END
316 
317 // y = b1 * exp[b2/(x+b3)]  +  e
318 NIST_BEGIN(MGH10)
319   b[0] * exp(b[1] / (x + b[2]))
320 NIST_END
321 
322 // y = b1 + b2*exp[-x*b4] + b3*exp[-x*b5]
323 NIST_BEGIN(MGH17)
324   b[0] + b[1] * exp(-x * b[3]) + b[2] * exp(-x * b[4])
325 NIST_END
326 
327 // y = b1*(1-exp[-b2*x])  +  e
328 NIST_BEGIN(Misra1a)
329   b[0] * (T(1.0) - exp(-b[1] * x))
330 NIST_END
331 
332 // y = b1 * (1-(1+b2*x/2)**(-2))  +  e
333 NIST_BEGIN(Misra1b)
334   b[0] * (T(1.0) - T(1.0)/ ((T(1.0) + b[1] * x / 2.0) * (T(1.0) + b[1] * x / 2.0)))
335 NIST_END
336 
337 // y = b1 * (1-(1+2*b2*x)**(-.5))  +  e
338 NIST_BEGIN(Misra1c)
339   b[0] * (T(1.0) - pow(T(1.0) + T(2.0) * b[1] * x, -0.5))
340 NIST_END
341 
342 // y = b1*b2*x*((1+b2*x)**(-1))  +  e
343 NIST_BEGIN(Misra1d)
344   b[0] * b[1] * x / (T(1.0) + b[1] * x)
345 NIST_END
346 
347 const double kPi = 3.141592653589793238462643383279;
348 // pi = 3.141592653589793238462643383279E0
349 // y =  b1 - b2*x - arctan[b3/(x-b4)]/pi  +  e
350 NIST_BEGIN(Roszman1)
351   b[0] - b[1] * x - atan2(b[2], (x - b[3]))/T(kPi)
352 NIST_END
353 
354 // y = b1 / (1+exp[b2-b3*x])  +  e
355 NIST_BEGIN(Rat42)
356   b[0] / (T(1.0) + exp(b[1] - b[2] * x))
357 NIST_END
358 
359 // y = b1 / ((1+exp[b2-b3*x])**(1/b4))  +  e
360 NIST_BEGIN(Rat43)
361   b[0] / pow(T(1.0) + exp(b[1] - b[2] * x), T(1.0) / b[3])
362 NIST_END
363 
364 // y = (b1 + b2*x + b3*x**2 + b4*x**3) /
365 //    (1 + b5*x + b6*x**2 + b7*x**3)  +  e
366 NIST_BEGIN(Thurber)
367   (b[0] + b[1] * x + b[2] * x * x  + b[3] * x * x * x) /
368   (T(1.0) + b[4] * x + b[5] * x * x + b[6] * x * x * x)
369 NIST_END
370 
371 // y = b1 + b2*cos( 2*pi*x/12 ) + b3*sin( 2*pi*x/12 )
372 //        + b5*cos( 2*pi*x/b4 ) + b6*sin( 2*pi*x/b4 )
373 //        + b8*cos( 2*pi*x/b7 ) + b9*sin( 2*pi*x/b7 )  + e
374 NIST_BEGIN(ENSO)
375   b[0] + b[1] * cos(T(2.0 * kPi) * x / T(12.0)) +
376          b[2] * sin(T(2.0 * kPi) * x / T(12.0)) +
377          b[4] * cos(T(2.0 * kPi) * x / b[3]) +
378          b[5] * sin(T(2.0 * kPi) * x / b[3]) +
379          b[7] * cos(T(2.0 * kPi) * x / b[6]) +
380          b[8] * sin(T(2.0 * kPi) * x / b[6])
381 NIST_END
382 
383 // y = (b1/b2) * exp[-0.5*((x-b3)/b2)**2]  +  e
384 NIST_BEGIN(Eckerle4)
385   b[0] / b[1] * exp(T(-0.5) * pow((x - b[2])/b[1], 2))
386 NIST_END
387 
388 struct Nelson {
389  public:
Nelsonceres::examples::Nelson390   Nelson(const double* const x, const double* const y)
391       : x1_(x[0]), x2_(x[1]), y_(y[0]) {}
392 
393   template <typename T>
operator ()ceres::examples::Nelson394   bool operator()(const T* const b, T* residual) const {
395     // log[y] = b1 - b2*x1 * exp[-b3*x2]  +  e
396     residual[0] = T(log(y_)) - (b[0] - b[1] * T(x1_) * exp(-b[2] * T(x2_)));
397     return true;
398   }
399 
400  private:
401   double x1_;
402   double x2_;
403   double y_;
404 };
405 
406 template <typename Model, int num_residuals, int num_parameters>
RegressionDriver(const std::string & filename,const ceres::Solver::Options & options)407 int RegressionDriver(const std::string& filename,
408                      const ceres::Solver::Options& options) {
409   NISTProblem nist_problem(FLAGS_nist_data_dir + filename);
410   CHECK_EQ(num_residuals, nist_problem.response_size());
411   CHECK_EQ(num_parameters, nist_problem.num_parameters());
412 
413   Matrix predictor = nist_problem.predictor();
414   Matrix response = nist_problem.response();
415   Matrix final_parameters = nist_problem.final_parameters();
416 
417   printf("%s\n", filename.c_str());
418 
419   // Each NIST problem comes with multiple starting points, so we
420   // construct the problem from scratch for each case and solve it.
421   int num_success = 0;
422   for (int start = 0; start < nist_problem.num_starts(); ++start) {
423     Matrix initial_parameters = nist_problem.initial_parameters(start);
424 
425     ceres::Problem problem;
426     for (int i = 0; i < nist_problem.num_observations(); ++i) {
427       problem.AddResidualBlock(
428           new ceres::AutoDiffCostFunction<Model, num_residuals, num_parameters>(
429               new Model(predictor.data() + nist_problem.predictor_size() * i,
430                         response.data() + nist_problem.response_size() * i)),
431           NULL,
432           initial_parameters.data());
433     }
434 
435     ceres::Solver::Summary summary;
436     Solve(options, &problem, &summary);
437 
438     // Compute the LRE by comparing each component of the solution
439     // with the ground truth, and taking the minimum.
440     Matrix final_parameters = nist_problem.final_parameters();
441     const double kMaxNumSignificantDigits = 11;
442     double log_relative_error = kMaxNumSignificantDigits + 1;
443     for (int i = 0; i < num_parameters; ++i) {
444       const double tmp_lre =
445           -std::log10(std::fabs(final_parameters(i) - initial_parameters(i)) /
446                       std::fabs(final_parameters(i)));
447       // The maximum LRE is capped at 11 - the precision at which the
448       // ground truth is known.
449       //
450       // The minimum LRE is capped at 0 - no digits match between the
451       // computed solution and the ground truth.
452       log_relative_error =
453           std::min(log_relative_error,
454                    std::max(0.0, std::min(kMaxNumSignificantDigits, tmp_lre)));
455     }
456 
457     const int kMinNumMatchingDigits = 4;
458     if (log_relative_error >= kMinNumMatchingDigits) {
459       ++num_success;
460     }
461 
462     printf("start: %d status: %s lre: %4.1f initial cost: %e final cost:%e "
463            "certified cost: %e total iterations: %d\n",
464            start + 1,
465            log_relative_error < kMinNumMatchingDigits ? "FAILURE" : "SUCCESS",
466            log_relative_error,
467            summary.initial_cost,
468            summary.final_cost,
469            nist_problem.certified_cost(),
470            (summary.num_successful_steps + summary.num_unsuccessful_steps));
471   }
472   return num_success;
473 }
474 
SetMinimizerOptions(ceres::Solver::Options * options)475 void SetMinimizerOptions(ceres::Solver::Options* options) {
476   CHECK(ceres::StringToMinimizerType(FLAGS_minimizer,
477                                      &options->minimizer_type));
478   CHECK(ceres::StringToLinearSolverType(FLAGS_linear_solver,
479                                         &options->linear_solver_type));
480   CHECK(ceres::StringToPreconditionerType(FLAGS_preconditioner,
481                                           &options->preconditioner_type));
482   CHECK(ceres::StringToTrustRegionStrategyType(
483             FLAGS_trust_region_strategy,
484             &options->trust_region_strategy_type));
485   CHECK(ceres::StringToDoglegType(FLAGS_dogleg, &options->dogleg_type));
486   CHECK(ceres::StringToLineSearchDirectionType(
487       FLAGS_line_search_direction,
488       &options->line_search_direction_type));
489   CHECK(ceres::StringToLineSearchType(FLAGS_line_search,
490                                       &options->line_search_type));
491   CHECK(ceres::StringToLineSearchInterpolationType(
492       FLAGS_line_search_interpolation,
493       &options->line_search_interpolation_type));
494 
495   options->max_num_iterations = FLAGS_num_iterations;
496   options->use_nonmonotonic_steps = FLAGS_nonmonotonic_steps;
497   options->initial_trust_region_radius = FLAGS_initial_trust_region_radius;
498   options->max_lbfgs_rank = FLAGS_lbfgs_rank;
499   options->line_search_sufficient_function_decrease = FLAGS_sufficient_decrease;
500   options->line_search_sufficient_curvature_decrease =
501       FLAGS_sufficient_curvature_decrease;
502   options->max_num_line_search_step_size_iterations =
503       FLAGS_max_line_search_iterations;
504   options->max_num_line_search_direction_restarts =
505       FLAGS_max_line_search_restarts;
506   options->use_approximate_eigenvalue_bfgs_scaling =
507       FLAGS_approximate_eigenvalue_bfgs_scaling;
508   options->function_tolerance = 1e-18;
509   options->gradient_tolerance = 1e-18;
510   options->parameter_tolerance = 1e-18;
511 }
512 
SolveNISTProblems()513 void SolveNISTProblems() {
514   if (FLAGS_nist_data_dir.empty()) {
515     LOG(FATAL) << "Must specify the directory containing the NIST problems";
516   }
517 
518   ceres::Solver::Options options;
519   SetMinimizerOptions(&options);
520 
521   std::cout << "Lower Difficulty\n";
522   int easy_success = 0;
523   easy_success += RegressionDriver<Misra1a,  1, 2>("Misra1a.dat",  options);
524   easy_success += RegressionDriver<Chwirut,  1, 3>("Chwirut1.dat", options);
525   easy_success += RegressionDriver<Chwirut,  1, 3>("Chwirut2.dat", options);
526   easy_success += RegressionDriver<Lanczos,  1, 6>("Lanczos3.dat", options);
527   easy_success += RegressionDriver<Gauss,    1, 8>("Gauss1.dat",   options);
528   easy_success += RegressionDriver<Gauss,    1, 8>("Gauss2.dat",   options);
529   easy_success += RegressionDriver<DanWood,  1, 2>("DanWood.dat",  options);
530   easy_success += RegressionDriver<Misra1b,  1, 2>("Misra1b.dat",  options);
531 
532   std::cout << "\nMedium Difficulty\n";
533   int medium_success = 0;
534   medium_success += RegressionDriver<Kirby2,   1, 5>("Kirby2.dat",   options);
535   medium_success += RegressionDriver<Hahn1,    1, 7>("Hahn1.dat",    options);
536   medium_success += RegressionDriver<Nelson,   1, 3>("Nelson.dat",   options);
537   medium_success += RegressionDriver<MGH17,    1, 5>("MGH17.dat",    options);
538   medium_success += RegressionDriver<Lanczos,  1, 6>("Lanczos1.dat", options);
539   medium_success += RegressionDriver<Lanczos,  1, 6>("Lanczos2.dat", options);
540   medium_success += RegressionDriver<Gauss,    1, 8>("Gauss3.dat",   options);
541   medium_success += RegressionDriver<Misra1c,  1, 2>("Misra1c.dat",  options);
542   medium_success += RegressionDriver<Misra1d,  1, 2>("Misra1d.dat",  options);
543   medium_success += RegressionDriver<Roszman1, 1, 4>("Roszman1.dat", options);
544   medium_success += RegressionDriver<ENSO,     1, 9>("ENSO.dat",     options);
545 
546   std::cout << "\nHigher Difficulty\n";
547   int hard_success = 0;
548   hard_success += RegressionDriver<MGH09,    1, 4>("MGH09.dat",    options);
549   hard_success += RegressionDriver<Thurber,  1, 7>("Thurber.dat",  options);
550   hard_success += RegressionDriver<BoxBOD,   1, 2>("BoxBOD.dat",   options);
551   hard_success += RegressionDriver<Rat42,    1, 3>("Rat42.dat",    options);
552   hard_success += RegressionDriver<MGH10,    1, 3>("MGH10.dat",    options);
553 
554   hard_success += RegressionDriver<Eckerle4, 1, 3>("Eckerle4.dat", options);
555   hard_success += RegressionDriver<Rat43,    1, 4>("Rat43.dat",    options);
556   hard_success += RegressionDriver<Bennet5,  1, 3>("Bennett5.dat", options);
557 
558   std::cout << "\n";
559   std::cout << "Easy    : " << easy_success << "/16\n";
560   std::cout << "Medium  : " << medium_success << "/22\n";
561   std::cout << "Hard    : " << hard_success << "/16\n";
562   std::cout << "Total   : " << easy_success + medium_success + hard_success << "/54\n";
563 }
564 
565 }  // namespace examples
566 }  // namespace ceres
567 
main(int argc,char ** argv)568 int main(int argc, char** argv) {
569   google::ParseCommandLineFlags(&argc, &argv, true);
570   google::InitGoogleLogging(argv[0]);
571   ceres::examples::SolveNISTProblems();
572   return 0;
573 };
574