| 368 | // ---------------------------------------------------------------------------------------- |
| 369 | |
| 370 | void test_anomaly_detection ( |
| 371 | ) |
| 372 | { |
| 373 | dlog << LINFO << " begin test_anomaly_detection()"; |
| 374 | // Here we declare that our samples will be 2 dimensional column vectors. |
| 375 | typedef matrix<double,2,1> sample_type; |
| 376 | |
| 377 | // Now we are making a typedef for the kind of kernel we want to use. I picked the |
| 378 | // radial basis kernel because it only has one parameter and generally gives good |
| 379 | // results without much fiddling. |
| 380 | typedef radial_basis_kernel<sample_type> kernel_type; |
| 381 | |
| 382 | // Here we declare an instance of the kcentroid object. The first argument to the constructor |
| 383 | // is the kernel we wish to use. The second is a parameter that determines the numerical |
| 384 | // accuracy with which the object will perform part of the learning algorithm. Generally |
| 385 | // smaller values give better results but cause the algorithm to run slower. You just have |
| 386 | // to play with it to decide what balance of speed and accuracy is right for your problem. |
| 387 | // Here we have set it to 0.01. |
| 388 | kcentroid<kernel_type> test(kernel_type(0.1),0.01); |
| 389 | |
| 390 | |
| 391 | svm_one_class_trainer<kernel_type> one_class_trainer; |
| 392 | one_class_trainer.set_nu(0.4); |
| 393 | one_class_trainer.set_kernel(kernel_type(0.2)); |
| 394 | |
| 395 | std::vector<sample_type> samples; |
| 396 | |
| 397 | // now we train our object on a few samples of the sinc function. |
| 398 | sample_type m; |
| 399 | for (double x = -15; x <= 8; x += 1) |
| 400 | { |
| 401 | m(0) = x; |
| 402 | m(1) = sinc(x); |
| 403 | test.train(m); |
| 404 | samples.push_back(m); |
| 405 | } |
| 406 | |
| 407 | decision_function<kernel_type> df = one_class_trainer.train(samples); |
| 408 | |
| 409 | running_stats<double> rs; |
| 410 | |
| 411 | // Now lets output the distance from the centroid to some points that are from the sinc function. |
| 412 | // These numbers should all be similar. We will also calculate the statistics of these numbers |
| 413 | // by accumulating them into the running_stats object called rs. This will let us easily |
| 414 | // find the mean and standard deviation of the distances for use below. |
| 415 | dlog << LDEBUG << "Points that are on the sinc function:\n"; |
| 416 | m(0) = -1.5; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 417 | m(0) = -1.5; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 418 | m(0) = -0; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 419 | m(0) = -0.5; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 420 | m(0) = -4.1; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 421 | m(0) = -1.5; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 422 | m(0) = -0.5; m(1) = sinc(m(0)); dlog << LDEBUG << " " << test(m); rs.add(test(m)); |
| 423 | |
| 424 | m(0) = -1.5; m(1) = sinc(m(0)); DLIB_TEST_MSG(rs.scale(test(m)) < 2, rs.scale(test(m))); |
| 425 | m(0) = -1.5; m(1) = sinc(m(0)); DLIB_TEST_MSG(rs.scale(test(m)) < 2, rs.scale(test(m))); |
| 426 | m(0) = -0; m(1) = sinc(m(0)); DLIB_TEST_MSG(rs.scale(test(m)) < 2, rs.scale(test(m))); |
| 427 | m(0) = -0.5; m(1) = sinc(m(0)); DLIB_TEST_MSG(rs.scale(test(m)) < 2, rs.scale(test(m))); |
no test coverage detected