1. Ruben Martinez
  2. BayesOpt
Issue #2 resolved

Problem with LHS in Python interface

kiudee
created an issue

(Copying this from Github, since you mentioned you prefer Bitbucket:)

bayesopt when using latin hypercube sampling seems to invoke intermediate callbacks resulting in more initial samples than intented.

Example input values received in callback function:

x = [ 0.30424932  0.          0.        ]
x = [ 0.30424932  0.32077927  0.        ]
x = [ 0.30424932  0.32077927  0.43445221]

It seems like it is filling the list one by one and invoking the callback in between.

Comments (2)

  1. kiudee reporter

    Addition:

    This happens for every evaluation of the target function. Here is an excerpt of demo_quad.py with an added print statement in testfunc:

    - 11:30:32.307268 INFO: Iteration: 2 of 50 | Total samples: 22
    - 11:30:32.307384 INFO: Query: [5](2.45692e-15,1,1,7.31385e-15,7.31385e-15)
    - 11:30:32.307432 INFO: Query outcome: 6.2245
    - 11:30:32.307464 INFO: Best query: [5](0.149761,0.708905,0.511304,0.554632,0.442278)
    - 11:30:32.307581 INFO: Best outcome: 5.27199
    [ 0.29410885  0.          0.          0.          0.        ]
    [ 0.29410885  0.49465416  0.          0.          0.        ]
    [ 0.29410885  0.49465416  0.22084766  0.          0.        ]
    [ 0.29410885  0.49465416  0.22084766  0.41855352  0.        ]
    [ 0.29410885  0.49465416  0.22084766  0.41855352  0.29884503]
    - 11:30:32.320429 INFO: Iteration: 3 of 50 | Total samples: 23
    - 11:30:32.320524 INFO: Query: [5](0.294109,0.494654,0.220848,0.418554,0.298845)
    - 11:30:32.320729 INFO: Query outcome: 5.04913
    - 11:30:32.320810 INFO: Best query: [5](0.294109,0.494654,0.220848,0.418554,0.298845)
    - 11:30:32.320994 INFO: Best outcome: 5.04913
    
  2. Log in to comment