I tried your example 'tutorial_interactive_mode.ipynb'.
This code save 'policy.predictor'object after bayes_search as
with open('predictor.dump', 'wb') as f:
pickle.dump(policy.predictor, f)
Then delete policy as
del policy
Then initialise policy again and read saved 'policy.predictor' object as
policy = combo.search.discrete.policy(test_X=X)
policy.load('history.npz', 'training.npz', 'predictor.dump')
Since 'predictor' object seems still alive in this code, I remake 'predicter' as
predictor=policy.predictor
and run
actions = policy.bayes_search(max_num_probes=1, num_search_each_probe=10,
predictor=predictor,
simulator=None, score='EI', interval=0, num_rand_basis = 0)
Resut is
UnboundLocalError: local variable 'grad_marlik' referenced before assignmenterror.
What is wrong ?
Just I want to run COMBO only adding our new experimental data,
since I think necessary prior information are saved in the policy.object of COMBO .
I tried your example 'tutorial_interactive_mode.ipynb'.
This code save 'policy.predictor'object after bayes_search as
with open('predictor.dump', 'wb') as f:
pickle.dump(policy.predictor, f)
Then delete policy as
del policy
Then initialise policy again and read saved 'policy.predictor' object as
policy = combo.search.discrete.policy(test_X=X)
policy.load('history.npz', 'training.npz', 'predictor.dump')
Since 'predictor' object seems still alive in this code, I remake 'predicter' as
predictor=policy.predictor
and run
actions = policy.bayes_search(max_num_probes=1, num_search_each_probe=10,
predictor=predictor,
simulator=None, score='EI', interval=0, num_rand_basis = 0)
Resut is
UnboundLocalError: local variable 'grad_marlik' referenced before assignmenterror.
What is wrong ?
Just I want to run COMBO only adding our new experimental data,
since I think necessary prior information are saved in the policy.object of COMBO .