- Please install a MKL library in /opt/intel/
- Type: source /opt/intel/mkl/bin/mklvars.sh intel64
- Go to ./blas_ver directory. Non-blas version is not available at this moment.
- Type: make
learn_curve_krls.cpp file simulates the learning accuracy of three different arithmetic KRLS.
- The execution command lines are:
./learn_krls_curve arithmetic_type appl_type min_#samples max#_samples input_length kernel_width ald_param
e.g., ./learn_krls_curve mixed mg30 1 10000 6 0.5 0.001
or
./learn_krls_curve double sunspot 120 2880 3 95 0.0001 (IN THIS application, traning data will have a stride of 120, should be fixed for input length to 3)
Run
./learn_krls_curve prec_type(double or mixed or single) application_type(sinc_lin_tri or mg30 or sunspot) first_#_samples(5) last_#_samples(50000) feature_size(3)
kernel_width(4.0) ald_threshold(0.001)
=>
#_tr_samples log10(RMSE) training_time testing_time D size
5 -0.34563 0.00619197 0.00875616 5
50 -0.611389 0.00492191 0.012069 42
500 -1.34673 0.0445101 0.012876 65
5000 -1.83638 0.21237 0.012913 66
50000 -2.1564 5.63238 0.0156641 68
NOTE: Do not put a fractional number as an argument.