ernstklrb
commented
4 years ago This error is a result of the acquisition_type being inconsistent with the model_type. When you just change model_type to 'GP_MCMC' it will fail with the default acquisition_type 'EI'. You also have to change the acquisition_type to, e.g., 'EI_MCMC'.
I find the library could and should automatically switch to the appropriate non-MCMC or MCMC version of an acquisition function.
Dear Javier, currently I am working on an optimization of hyperparameters for solving the traveling salesman problem with a heuristic for my masterthesis. Therefore I am using GPyOpt. The normal GP is working. I tried to use GP_MCMC for a better fit of the GP hyperparameters with these settings:
myBopt = BayesianOptimization(f=Optimization.RadiusFunction, model_type='GP_MCMC', acquistion_type='EI_MCMC', acquisition_optimizer_type='lbfgs', normalize_Y=True, n_samples=5, domain=domain, exact_feval=False) myBopt.run_optimization(max_iter=1250, max_time=None, verbosity=True, evaluations_file='eval.txt', report_file='report.txt', models_file='models.txt')
Resulting in this error:
reconstraining parameters GP_regression.rbf reconstraining parameters GP_regression.Gaussian_noise.variance Traceback (most recent call last): File "C:/Users/Duc Hoang/Desktop/Masterarbeit/Optimization", line 96, in
models_file='models.txt')
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\core\bo.py", line 143, in run_optimization
self.suggested_sample = self._compute_next_evaluations()
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\core\bo.py", line 234, in _compute_next_evaluations
return self.space.zip_inputs(self.evaluator.compute_batch(duplicate_manager=duplicate_manager, context_manager= self.acquisition.optimizer.context_manager))
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\core\evaluators\sequential.py", line 22, in computebatch
x, = self.acquisition.optimize(duplicate_manager=duplicate_manager)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\acquisitions\base.py", line 59, in optimize
out = self.optimizer.optimize(f=self.acquisition_function, f_df=self.acquisition_function_withGradients, duplicate_manager=duplicate_manager)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\optimization\acquisition_optimizer.py", line 69, in optimize
anchor_points = anchor_points_generator.get(duplicate_manager=duplicate_manager, context_manager=self.context_manager)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\optimization\anchor_points_generator.py", line 59, in get
scores = self.get_anchor_point_scores(X)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\optimization\anchor_points_generator.py", line 98, in get_anchor_point_scores
return self.objective(X).flatten()
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\acquisitions\base.py", line 37, in acquisition_function
f_acqu = self._compute_acq(x)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\acquisitions\EI.py", line 38, in _compute_acq
phi, Phi, u = get_quantiles(self.jitter, fmin, m, s)
File "C:\Users\Duc Hoang\AppData\Local\Programs\Python\Python36\lib\site-packages\GPyOpt\util\general.py", line 123, in get_quantiles
elif s< 1e-10:
TypeError: '<' not supported between instances of 'list' and 'float'
In addition, I am using these package versions:
GPy 1.9.6 GPyOpt 1.2.5 matplotlib 3.0.3 numpy 1.15.4 paramz 0.9.4 scipy 1.2.1
with Python 3.6. Thanks in advance!