r - Hyper-parameter optimization in SVM using Bayesian Optimization -


i trying search optimal values of c , gamma in svm (with gaussian kernel) using bayesian optimization (instead of grid-search) technique. end, using svm_opt function in r (details of function usage can found here):

svm_opt(train_data, train_label, test_data, test_label,   gamma_range = c(10^(-5), 10^5), c_range = c(10^(-2), 10^2),   init_points = 20, n_iter = 1, acq = "ucb", kappa = 2.576, eps = 0,   kernel = list(type = "exponential", power = 2), init_points=21) 

however, running following error:

error in gp_deviance(param_init_200d[i, ], x, y, nug_thres, corr = corr) :  infinite values of deviance function, unable find optimum parameters 

it error related gp_deviance have tried overcome tweaking the:

  1. search range (by setting gamma_range = c(2^(-3), 2^3) , c_range = c(2^(-3), 2^3)),
  2. tunable parameters eps , kappa (gp upper confidence bound , expected improvement respectively) determine trade-off between exploration , exploitation,
  3. acquisition function type, acq
  4. values of init_points

none of these have been successful, , after init_points-1 points aforementioned error occurs.

is there way conduct search completion? advice , insights helpful.


Comments

Popular posts from this blog

android - InAppBilling registering BroadcastReceiver in AndroidManifest -

python Tkinter Capturing keyboard events save as one single string -

sql server - Why does Linq-to-SQL add unnecessary COUNT()? -