| 3963 | // ---------------------------------------------------------------------------------------- |
| 3964 | |
| 3965 | void test_simple_linear_regression_with_mult_prev() |
| 3966 | { |
| 3967 | srand(1234); |
| 3968 | print_spinner(); |
| 3969 | const int num_samples = 1000; |
| 3970 | ::std::vector<matrix<double>> x(num_samples); |
| 3971 | ::std::vector<float> y(num_samples); |
| 3972 | const float true_slope = 2.0; |
| 3973 | for ( int ii = 0; ii < num_samples; ++ii ) |
| 3974 | { |
| 3975 | const double val = static_cast<double>(ii-500)/100; |
| 3976 | matrix<double> tmp(1,1); |
| 3977 | tmp = val; |
| 3978 | x[ii] = tmp; |
| 3979 | y[ii] = ( true_slope*static_cast<float>(val*val)); |
| 3980 | } |
| 3981 | |
| 3982 | randomize_samples(x,y); |
| 3983 | |
| 3984 | using net_type = loss_mean_squared<fc<1, mult_prev1<fc<2,tag1<fc<2,input<matrix<double>>>>>>>>; |
| 3985 | net_type net; |
| 3986 | sgd defsolver(0,0.9); |
| 3987 | dnn_trainer<net_type> trainer(net, defsolver); |
| 3988 | trainer.set_learning_rate(1e-5); |
| 3989 | trainer.set_min_learning_rate(1e-11); |
| 3990 | trainer.set_mini_batch_size(50); |
| 3991 | trainer.set_max_num_epochs(2000); |
| 3992 | trainer.train(x, y); |
| 3993 | |
| 3994 | running_stats<double> rs; |
| 3995 | for (size_t i = 0; i < x.size(); ++i) |
| 3996 | { |
| 3997 | double val = y[i]; |
| 3998 | double out = net(x[i]); |
| 3999 | rs.add(std::abs(val-out)); |
| 4000 | } |
| 4001 | dlog << LINFO << "rs.mean(): " << rs.mean(); |
| 4002 | dlog << LINFO << "rs.stddev(): " << rs.stddev(); |
| 4003 | dlog << LINFO << "rs.max(): " << rs.max(); |
| 4004 | DLIB_TEST(rs.mean() < 0.1); |
| 4005 | } |
| 4006 | |
| 4007 | // ---------------------------------------------------------------------------------------- |
| 4008 |
no test coverage detected