4. Code Overview

4.1. Modules in COMBO

Instruction to import the modules in COMBO:

Relative import 'gp', should be 'combo.gp' (relative-import)
Relative import 'opt', should be 'combo.opt' (relative-import)
Relative import 'blm', should be 'combo.blm' (relative-import)
Relative import 'misc', should be 'combo.misc' (relative-import)
Relative import 'search', should be 'combo.search' (relative-import)
Relative import 'predictor', should be 'combo.predictor' (relative-import)
Relative import 'variable', should be 'combo.variable' (relative-import)

4.2. Structure of COMBO

The files in each directory:

|-- blm
|   |-- basis
|   |   |-- fourier.py
|   |   `-- __init__.py
|   |-- core
|   |   |-- __init__.py
|   |   `-- model.py
|   |-- inf
|   |   |-- exact.py
|   |   `-- __init__.py
|   |-- __init__.py
|   |-- lik
|   |   |-- gauss.py
|   |   |-- __init__.py
|   |   |-- linear.py
|   |   `-- _src
|   |       |-- cov.py
|   |       `-- __init__.py
|   |-- predictor.py
|   `-- prior
|       |-- gauss.py
|       `-- __init__.py
|-- gp
|   |-- core
|   |   |-- __init__.py
|   |   |-- learning.py
|   |   |-- model.py
|   |   `-- prior.py
|   |-- cov
|   |   |-- gauss.py
|   |   |-- gauss.pyc
|   |   |-- __init__.py
|   |   `-- _src
|   |       |-- enhance_gauss.c
|   |       |-- enhance_gauss.pyx
|   |       |-- __init__.py
|   |       `-- __init__.pyc
|   |-- inf
|   |   |-- exact.py
|   |   `-- __init__.py
|   |-- __init__.py
|   |-- lik
|   |   |-- gauss.py
|   |   `-- __init__.py
|   |-- mean
|   |   |-- const.py
|   |   |-- __init__.py
|   |   `-- zero.py
|   `-- predictor.py
|-- __init__.py
|-- misc
|   |-- centering.py
|   |-- gauss_elim.py
|   |-- __init__.py
|   |-- set_config.py
|   `-- _src
|       |-- cholupdate.c
|       |-- cholupdate.pyx
|       |-- diagAB.c
|       |-- diagAB.pyx
|       |-- __init__.py
|       |-- logsumexp.c
|       |-- logsumexp.pyx
|       |-- traceAB.c
|       `-- traceAB.pyx
|-- opt
|   |-- adam.py
|   `-- __init__.py
|-- predictor.py
|-- search
|   |-- call_simulator.py
|   |-- discrete
|   |   |-- __init__.py
|   |   |-- policy.py
|   |   `-- results.py
|   |-- __init__.py
|   |-- score.py
|   `-- utility.py
`-- variable.py

4.3. Classes and functions in COMBO

4.3.1. variable object

COMBO defines a python class called variable to setup and handle the dataset in the training and testing. From a python script, variable object can be created like this:

>>>from combo.variable import variable
>>>import numpy as np
>>>X= np.array([[1,2,3,4], [5,6,7,8], [9,10,11,12]])
>>>t=np.array([[1], [5], [9]])
>>>Z=np.array([[2,6,3,4], [9,6,10,8], [22,10,11,12]])
>>>vtest=variable(X,t,Z)
>>>print vtest.X
class combo.variable.variable
X
array_like
t
array_like
Z
array_like
get_subset(index)
delete(num_row)
add(X, t, Z)
delete_X(num_row)
delete_t(num_row)
delete_Z(num_row)
add_X(X)
add_t(t)
add_Z(Z)
save(filename)
load(filename)
class combo.predictor.base_predictor
config
model
fit()
prepare()
delete_stats()
get_basis()
get_post_fmean()
get_post_fcov()
get_post_params()
get_post_samples()
combo.misc.centering(X)

standardize the data X and remove the columns with zero value

\[X_{1,1\sigma}=\frac{X_{i}-\bar{X}_S}{\sigma_{X,S}}\]
Parameters:X – input data
Return type:return the standardized data for the columns with non-zero values
combo.misc.gauss_elim(L, t)
Parameters:
  • L –
  • t –
Return type:

return
class combo.misc.set_config.set_config
show()
load()
class combo.misc.set_config.search
show()
load()
class combo.misc.set_config.learning
show()
load()
class combo.misc.set_config.batch
show()
load()
class combo.misc.set_config.online
show()
load()
class combo.misc.set_config.adam
show()
load()
combo.misc.set_config.boolean(str)
Parameters:str –
Return type:return a boolean value
combo.search.utility.show_search_results(history, N)
Parameters:
  • history –
  • N –
Parameters:
  • N –
  • score –
combo.search.utility.show_interactive_mode(simulator, history)
Parameters:
  • simulator –
  • history –
combo.search.utility.is_learning(n, interval)
Parameters:
  • n –
  • interval –
combo.search.call_simulator(simu, action)
Parameters:
  • simu –
  • action –
combo.search.score.EI(predictor, training, test, fmax=None)
Parameters:
  • predictor –
  • training –
  • test –
  • fmax –
Return type:

return score
combo.search.score.PI(predictor, training, test, fmax=None)
Parameters:
  • predictor –
  • training –
  • test –
  • fmax –
Return type:

return score
combo.search.score.TS(predictor, training, test, alpha=1)
Parameters:
  • predictor –
  • training –
  • test –
  • alpha –
Return type:

return score
class combo.search.discrete.policy.policy
set_seed()
delete_actions()
write()
get_score()
get_marginal_score()
get_actions()
get_random_action()
load()
export_predictor()
export_training()
export_history()
_set_predictor()
_init_predictor()
_set_training()
_set_unchosed_actions()
_set_test()
_set_config()
class combo.search.discrete.results.history
write()
export_sequence_best_fx()
export_all_sequence_best_fx()
save()
load()
class combo.opt.adam
set_params()
update()
run()
_set_options()
class combo.gp.predictor.predictor
fit()
get_basis()
get_post_params()
prepare()
delete_stats()
get_post_fmean()
get_post_fcov()
get_post_samples()
get_predict_samples()
class combo.gp.core.learning.batch
run()
one_run()
class combo.gp.core.learning.online
run()
one_run()
disp_marlik()
get_one_update()
class combo.gp.core.learning.adam
rest()
get_one_update()
class combo.gp.core.model.model
cat_params()
decomp_params()
set_params()
sub_sampling()
export_blm()
eval_marlik()
get_grad_marlik()
get_params_bound()
prepare()
get_post_fmean()
get_post_fcov()
post_sampling()
predict_sampling()
print_params()
get_cand_params()
fit()
class combo.gp.core.prior
cat_params()
decomp_params()
get_mean()
get_cov()
get_grad_mean()
get_grad_cov()
set_params()
set_mean_params()
set_cov_params()
sampling()
class combo.gp.cov.gauss.gauss
cat_params()
print_params()
prepare()
get_grad()
get_cov()
set_params()
supp_params()
decomp_params()
save()
load()
get_params_bound()
cat_params()
rand_expans()
get_cand_params()
combo.gp.inf.exact.eval_marlik(gp, X, t, params = None)
Parameters:
  • gp –
  • X –
  • t –
  • params –
Return type:

return marlik
combo.gp.inf.exact.get_grad_marlik(gp, X, t, params = None)
Parameters:
  • gp –
  • X –
  • t –
  • params –
Return type:

return grad_marlik
combo.gp.inf.exact.prepare(gp, X, t, params = None)
Parameters:
  • gp –
  • X –
  • t –
  • params –
Return type:

return stats
combo.gp.inf.get_post_fmean(gp, X, t, params = None)
Parameters:
  • gp –
  • X –
  • t –
  • params –
Return type:

return G.dot(alpha) + fmu
combo.gp.inf.get_post_fcov(gp, X, t, params = None, diag = True)
Parameters:
  • gp –
  • X –
  • t –
  • params –
  • diag –
Return type:

post_cov
class combo.gp.lik.gauss.gauss
supp_params()
trans_params()
get_params_bound()
get_cov()
get_grad()
set_params()
get_cand_params()
sampling()
class combo.gp.mean.const.const
supp_params()
get_params_bound()
get_mean()
get_grad()
set_params()
init_params()
get_cand_params()
class combo.blm.predictor.predictor
fit()
prepare()
delete_stats()
get_basis()
get_post_fmean()
get_post_fcov()
get_post_params()
get_post_samples()
get_predict_samples()
update()
class combo.blm.basis.fourier
get_basis()
set_params()
show()
_check_params()
_check_len_params()
class combo.blm.core.model.model
prepare()
update_stats()
get_post_params_mean()
get_post_fmean()
sampling()
post_sampling()
predict_sampling()
get_post_fcov()
_set_options()
_init_prior()
combo.blm.inf.exact.prepare(blm, X, t, Psi = None)
Parameters:
  • blm –
  • X –
  • t –
  • Psi –
combo.blm.inf.exact.update_stats(blm, x, t, Psi = None)
Parameters:
  • blm –
  • x –
  • t –
  • Psi –
Return type:

return ( L, b, alpha )
combo.blm.inf.exact.sampling(blm, w_mu = None, N=1, alpha = 1.0)
Parameters:
  • blm –
  • w_mu –
  • N –
  • alpha –
Return type:

return (invLz.transpose() + w_mu).transpose()
combo.blm.inf.exact.get_post_params_mean(blm)
Parameters:blm –
Return type:return blm.stats[2] * blm.lik.cov.prec
combo.blm.inf.exact.get_post_fmean(blm, X, Psi = None, w = None)
Parameters:
  • blm –
  • X –
  • Psi –
  • w –
Return type:

return Psi.dot(w) + blm.lik.linear.bias
combo.blm.inf.exact.get_post_fcov(blm, X, Psi = None, diag = True)
Parameters:
  • blm –
  • X –
  • Psi –
  • diag –
Return type:

return fcov
class combo.blm.lik.gauss.gauss
get_cov()
get_prec()
get_basis()
get_mean()
set_params()
set_bias()
sampling()
class combo.blm.lik.linear.linear
get_mean()
set_params()
set_bias()
_init_params()
_init_bias()
class combo.blm.prior.gauss.cov_const
get_cov()
get_prec()
set_params()
_trans_params()
class combo.blm.prior.gauss.gauss
get_mean()
get_cov()
get_prec()
set_params()
_init_cov()