2 stage Gaussian Process simulation by Mont Carlo

Main folder:

• script: (TEST: An Experiment of a simulation; test: an experiments of a function)
  • TEST_data_generation: Generate synthetic data by known GPs
  • TEST_MAIN: sample by annealed Importance Sampling
  • TEST_SA: find peaks by Simulated Annealing, and then sample by M-H
  • TEST_plot_hiD: Try to plot high dimrensional posterior
  • TEST_Gibbs: sample by Gibbs sampling
  • test_congau: test conditional Gaussian distribution function
• functions
  • kfcn: Gaussian kernel function
  • pos_bond: convert range (-inf,inf) to [0,positive number]
  • logmvnpdf: log pdf of multivariabel Gaussian distribution
  • mvhist: plot histogram of high dimensional samples
  • resample: resample the same number of samples by weights
  • Ly_Given_z: likelihood of z by observed output y
  • Pz_Given_x: probability of z given observed inpit x
  • my_fitrgp: Gaussian Process Regression by given kernel
  • congau: find parameters of conditional Gaussian distribution
• data file
  • data: generated data by TEST_data_generation
• figure
  • figure of TEST_MAIN
    • 6 IOs across 10 seconds
      • Pred_y_ALL_z: prediction of y using all samples
      • Pred_y_Half_z: prediction of y using samples that z1 > z2
    • 6 IOs across 5 seconds
      • Pred_y_ALL_z2: prediction of y using all samples
      • Pred_y_Half_z2: prediction of y using samples that z1 > z2
    • 6 IOs across 5 seconds, using all samples, but learn with fake kernels
      • Pred_y_too_narrow: using kernels that are narrower than ground truth
      • Pred_y_too_wide: using kernels that are wider than ground truth
  • figure of TEST_Gibbs
    • gibbs6: 6 IOs across 5 seconds
    • gibbs10: 11 IOs across 10 seconds

test the number of sections:

• script
  • test: find the growth of number of sections along with the groth of dimensions
• functions
  • section: recursive function to finding number of sections where d dimensional space divide by L hyperplanes
  • NS: find the number of sections in GP problem

history: archieved files that does not have much use