combo documentation - read the docs · return type return score combo.search.score.pi(predictor,...

combo DocumentationRelease 0.1a

Tsuda Laboratory

August 01, 2016

Contents

1 Introduction 31.1 Getting Help . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.2 Citation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31.3 Credits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.4 Licence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41.5 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4

2 Methodology 52.1 Generals in Bayesian optimization approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.2 Technical features in COMBO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52.3 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

3 Installation 73.1 Required Packages . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73.2 Install . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73.3 Uninstall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

4 Code Overview 94.1 Modules in COMBO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94.2 Structure of COMBO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94.3 Classes and functions in COMBO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

5 Tutorials 235.1 Random search . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235.2 Bayesian optimization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 Indices and tables 25

Bibliography 27

i

combo Documentation, Release 0.1a

Contents:

Contents 1


2 Contents

CHAPTER 1

Introduction

COMmon Bayesian Optimization Library (COMBO)

Bayesian optimization [Moc74] has been proven as an effective tool in accelerating scientific discovery. A standardimplementation (e.g., scikit-learn), however, can accommodate only small training data. COMBO [URH+16] is highlyscalable due to an efficient protocol that employs Thompson sampling [CL11], random feature maps [RR08], one-rankCholesky update [GGMS72] and automatic hyperparameter tuning [CER06].

1.1 Getting Help

The latest version of COMBO and documentation can always be found at https://github.com/tsudalab/combo.

1.2 Citation

We ask that you acknowledge the use of COMBO in any publications arising from the use of this code through thefollowing reference

[ref] Tsuyoshi Ueno, Trevor David Rhone, Zhufeng Hou, Teruyasu Mizoguchi and Koji Tsuda, COMBO: An Effi-cient Bayesian Optimization Library for Materials Science, Materials Discovery, (2016), in press. Available fromhttp://dx.doi.org/10.1016/j.md.2016.04.001

Bibtex file for citing COMBO

@article{Ueno2016,title = "COMBO: An Efficient Bayesian Optimization Library for Materials Science ",journal = "Materials Discovery",volume = "",number = "",pages = "-",year = "2016",note = "",doi = "http://dx.doi.org/10.1016/j.md.2016.04.001",url = "http://www.sciencedirect.com/science/article/pii/S2352924516300035",author = "Tsuyoshi Ueno and Trevor David Rhone and Zhufeng Hou and Teruyasu Mizoguchi and Koji Tsuda",}

3

http://scikit-learn.org/

https://github.com/tsudalab/combo

http://dx.doi.org/10.1016/j.md.2016.04.001


1.3 Credits

1.4 Licence

This package is distributed under the MIT License.

The MIT License (MIT).

Copyright (c) <2015-> <Tsuda Laboratory>

Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documen-tation files (the “Software”), to deal in the Software without restriction, including without limitation the rights to use,copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whomthe Software is furnished to do so, subject to the following conditions:

The above copyright notice and this permission notice shall be included in all copies or substantial portions of theSoftware.

THE SOFTWARE IS PROVIDED “AS IS”, WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED,INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PAR-TICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHTHOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTIONOF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFT-WARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.

1.5 References

4 Chapter 1. Introduction

https://en.wikipedia.org/wiki/MIT_License

https://opensource.org/licenses/MIT

http://www.tsudalab.org/

CHAPTER 2

Methodology

2.1 Generals in Bayesian optimization approach

∫︁ 𝑏

𝑎

𝑓 =

[︂𝑓(𝑎 +

𝑑𝑥

2) + 𝑓(𝑎 + 𝑑𝑥 +

𝑑𝑥

2) + (𝑓(𝑎 + 2𝑑𝑥 +

𝑑𝑥

2) + · · ·

]︂𝑑𝑥

2.1.1 Gaussian process

2.1.2 Acquisition Functions for Bayesian Optimization

2.2 Technical features in COMBO

2.2.1 Thompson sampling [CL11]

2.2.2 Random Feature Map [RR08]

2.2.3 Automatic hyperparameter tuning [CER06]

2.3 References

5


6 Chapter 2. Methodology

CHAPTER 3

Installation

3.1 Required Packages

• Python 2.7.x

• numpy >=1.10

• scipy >= 0.16

• Cython >= 0.22.1

• mpi4py >= 2.0 (optional)

1. Copy the following contents to a text file named ‘requirements.txt’

## To install these requirements, run## pip install -U -r requirements.txt## (the -U option also upgrades packages; from the second time on,## just run## pip install -r requirements.txt#### NOTE: before running the command above, you need to install a recent version## of pip from the website, and then possibly install/upgrade setuptools using## sudo pip install --upgrade setuptools## numpynumpy >=1.10

## scipyscipy >= 0.16

##Cython >= 0.22.1

## mpi4pympi4py >= 2.0 (optional)

2. Run the command

> pip install -U -r requirements.txt

3.2 Install

1. Download or clone the github repository, e.g.

7


> git clone https://github.com/tsudalab/combo.git

2. Run setup.py install

> cd combo> python setup.py install

3.3 Uninstall

1. Delete all installed files, e.g.

> python setup.py install --record file.txt> cat file.txt | xargs rm -rvf

8 Chapter 3. Installation


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 apython 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

Xarray_like

tarray_like

Zarray_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()

4.3. Classes and functions in COMBO 11


get_post_fcov()

get_post_params()

get_post_samples()

combo.misc.centering(X)standardize the data X and remove the columns with zero value

𝑋1,1𝜎 =𝑋𝑖 − �̄�𝑆

𝜎𝑋,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 –

combo.search.utility.show_start_message_multi_search(N, score=None)

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 –




combo.search.score.TS(predictor, training, test, alpha=1)

Parameters

• predictor –

• training –

• test –

• alpha –


class combo.search.discrete.policy.policy

set_seed()

delete_actions()

write()

random_search()

bayes_search()

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()

init_params_search()

class combo.gp.core.learning.online

run()

one_run()

disp_marlik()

init_params_search()

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()


CHAPTER 5

Tutorials

5.1 Random search

test.python

5.1.1 Multiple probe search

This is a normal text paragraph. The next paragraph is a code sample

5.1.2 Interactive mode search

5.2 Bayesian optimization

5.2.1 Multiple probe search

5.2.2 Interactive mode search

23


24 Chapter 5. Tutorials

CHAPTER 6

Indices and tables

• genindex

• modindex

• search

25


26 Chapter 6. Indices and tables

Bibliography

[CER06] Christopher K. I. Williams Carl Edward Rasmussen. Gaussian Processes for Machine Learn-ing. volume of Adaptive computation and machine learning. MIT Press, edition, 2006. URL:http://www.gaussianprocess.org/gpml/.

[CL11] Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling. In J. Shawe-Taylor, R. S.Zemel, P. L. Bartlett, F. Pereira, and K. Q. Weinberger, editors, Advances in Neural Information Processing Sys-tems 24, pages 2249–2257. Curran Associates, Inc., 2011. URL: http://papers.nips.cc/paper/4321-an-empirical-evaluation-of-thompson-sampling.pdf.

[GGMS72] Phillip E. Gill, Gene H. Golub, Walter A. Murray, and Michael A. Saunders. Methods for modify-ing matrix factorizations. Math. Comp., 28:505–535, 1972. URL: http://www.ams.org/journals/mcom/1974-28-126/S0025-5718-1974-0343558-6/, doi:http://dx.doi.org/10.1090/S0025-5718-1974-0343558-6.

[Moc74] Jonas Mockus. On bayesian methods for seeking the extremum. In Proceedings ofthe IFIP Technical Conference, 400–404. London, UK, UK, 1974. Springer-Verlag. URL:http://dl.acm.org/citation.cfm?id=646296.687872.

[RR08] Ali Rahimi and Benjamin Recht. Random features for large-scale kernel machines. In J. C. Platt, D. Koller,Y. Singer, and S. T. Roweis, editors, Advances in Neural Information Processing Systems 20, 1177–1184.Curran Associates, Inc., 2008. URL: http://papers.nips.cc/paper/3182-random-features-for-large-scale-kernel-machines.pdf.

[URH+16] Tsuyoshi Ueno, Trevor David Rhone, Zhufeng Hou, Teruyasu Mizoguchi, and KojiTsuda. Combo: an efficient bayesian optimization library for materials science. Materials Dis-covery, ():–, 2016. URL: http://www.sciencedirect.com/science/article/pii/S2352924516300035,doi:http://dx.doi.org/10.1016/j.md.2016.04.001.

[CER06] Christopher K. I. Williams Carl Edward Rasmussen. Gaussian Processes for Machine Learn-ing. volume of Adaptive computation and machine learning. MIT Press, edition, 2006. URL:http://www.gaussianprocess.org/gpml/.

[CL11] Olivier Chapelle and Lihong Li. An empirical evaluation of thompson sampling. In J. Shawe-Taylor, R. S.Zemel, P. L. Bartlett, F. Pereira, and K. Q. Weinberger, editors, Advances in Neural Information Processing Sys-tems 24, pages 2249–2257. Curran Associates, Inc., 2011. URL: http://papers.nips.cc/paper/4321-an-empirical-evaluation-of-thompson-sampling.pdf.

[GGMS72] Phillip E. Gill, Gene H. Golub, Walter A. Murray, and Michael A. Saunders. Methods for modify-ing matrix factorizations. Math. Comp., 28:505–535, 1972. URL: http://www.ams.org/journals/mcom/1974-28-126/S0025-5718-1974-0343558-6/, doi:http://dx.doi.org/10.1090/S0025-5718-1974-0343558-6.

27

http://www.gaussianprocess.org/gpml/

http://papers.nips.cc/paper/4321-an-empirical-evaluation-of-thompson-sampling.pdf


http://www.ams.org/journals/mcom/1974-28-126/S0025-5718-1974-0343558-6/


http://dx.doi.org/http://dx.doi.org/10.1090/S0025-5718-1974-0343558-6

http://dl.acm.org/citation.cfm?id=646296.687872

http://papers.nips.cc/paper/3182-random-features-for-large-scale-kernel-machines.pdf


http://www.sciencedirect.com/science/article/pii/S2352924516300035

http://dx.doi.org/http://dx.doi.org/10.1016/j.md.2016.04.001

http://www.gaussianprocess.org/gpml/





http://dx.doi.org/http://dx.doi.org/10.1090/S0025-5718-1974-0343558-6


[Moc74] Jonas Mockus. On bayesian methods for seeking the extremum. In Proceedings ofthe IFIP Technical Conference, 400–404. London, UK, UK, 1974. Springer-Verlag. URL:http://dl.acm.org/citation.cfm?id=646296.687872.

[RR08] Ali Rahimi and Benjamin Recht. Random features for large-scale kernel machines. In J. C. Platt, D. Koller,Y. Singer, and S. T. Roweis, editors, Advances in Neural Information Processing Systems 20, 1177–1184.Curran Associates, Inc., 2008. URL: http://papers.nips.cc/paper/3182-random-features-for-large-scale-kernel-machines.pdf.

[URH+16] Tsuyoshi Ueno, Trevor David Rhone, Zhufeng Hou, Teruyasu Mizoguchi, and KojiTsuda. Combo: an efficient bayesian optimization library for materials science. Materials Dis-covery, ():–, 2016. URL: http://www.sciencedirect.com/science/article/pii/S2352924516300035,doi:http://dx.doi.org/10.1016/j.md.2016.04.001.

28 Bibliography

http://dl.acm.org/citation.cfm?id=646296.687872



http://www.sciencedirect.com/science/article/pii/S2352924516300035

http://dx.doi.org/http://dx.doi.org/10.1016/j.md.2016.04.001

Index

Symbols_check_len_params() (combo.blm.basis.fourier method),

19_check_params() (combo.blm.basis.fourier method), 19_init_bias() (combo.blm.lik.linear.linear method), 20_init_cov() (combo.blm.prior.gauss.gauss method), 21_init_params() (combo.blm.lik.linear.linear method), 20_init_predictor() (combo.search.discrete.policy.policy

method), 14_init_prior() (combo.blm.core.model.model method), 19_set_config() (combo.search.discrete.policy.policy

method), 14_set_options() (combo.blm.core.model.model method),

19_set_options() (combo.opt.adam method), 15_set_predictor() (combo.search.discrete.policy.policy

method), 14_set_test() (combo.search.discrete.policy.policy method),

14_set_training() (combo.search.discrete.policy.policy

method), 14_set_unchosed_actions() (combo.search.discrete.policy.policy

method), 14_trans_params() (combo.blm.prior.gauss.cov_const

method), 20

Aadd() (combo.variable.variable method), 11add_t() (combo.variable.variable method), 11add_X() (combo.variable.variable method), 11add_Z() (combo.variable.variable method), 11array_like (combo.variable.variable attribute), 11

Bbayes_search() (combo.search.discrete.policy.policy

method), 14

Ccat_params() (combo.gp.core.model.model method), 15cat_params() (combo.gp.core.prior method), 16

cat_params() (combo.gp.cov.gauss.gauss method), 16combo.blm.basis.fourier (built-in class), 18combo.blm.core.model.model (built-in class), 19combo.blm.inf.exact.get_post_fcov() (built-in function),

20combo.blm.inf.exact.get_post_fmean() (built-in func-

tion), 20combo.blm.inf.exact.get_post_params_mean() (built-in

function), 19combo.blm.inf.exact.prepare() (built-in function), 19combo.blm.inf.exact.sampling() (built-in function), 19combo.blm.inf.exact.update_stats() (built-in function), 19combo.blm.lik.gauss.gauss (built-in class), 20combo.blm.lik.linear.linear (built-in class), 20combo.blm.predictor.predictor (built-in class), 18combo.blm.prior.gauss.cov_const (built-in class), 20combo.blm.prior.gauss.gauss (built-in class), 21combo.gp.core.learning.adam (built-in class), 15combo.gp.core.learning.batch (built-in class), 15combo.gp.core.learning.online (built-in class), 15combo.gp.core.model.model (built-in class), 15combo.gp.core.prior (built-in class), 16combo.gp.cov.gauss.gauss (built-in class), 16combo.gp.inf.exact.eval_marlik() (built-in function), 17combo.gp.inf.exact.get_grad_marlik() (built-in function),

17combo.gp.inf.exact.prepare() (built-in function), 17combo.gp.inf.get_post_fcov() (built-in function), 17combo.gp.inf.get_post_fmean() (built-in function), 17combo.gp.lik.gauss.gauss (built-in class), 18combo.gp.mean.const.const (built-in class), 18combo.gp.predictor.predictor (built-in class), 15combo.misc.centering() (built-in function), 12combo.misc.gauss_elim() (built-in function), 12combo.misc.set_config.adam (built-in class), 12combo.misc.set_config.batch (built-in class), 12combo.misc.set_config.boolean() (built-in function), 12combo.misc.set_config.learning (built-in class), 12combo.misc.set_config.online (built-in class), 12combo.misc.set_config.search (built-in class), 12combo.misc.set_config.set_config (built-in class), 12

29


combo.opt.adam (built-in class), 14combo.predictor.base_predictor (built-in class), 11combo.search.call_simulator() (built-in function), 13combo.search.discrete.policy.policy (built-in class), 14combo.search.discrete.results.history (built-in class), 14combo.search.score.EI() (built-in function), 13combo.search.score.PI() (built-in function), 13combo.search.score.TS() (built-in function), 13combo.search.utility.is_learning() (built-in function), 13combo.search.utility.show_interactive_mode() (built-in

function), 13combo.search.utility.show_search_results() (built-in

function), 13combo.search.utility.show_start_message_multi_search()

(built-in function), 13combo.variable.variable (built-in class), 11config (combo.predictor.base_predictor attribute), 11

Ddecomp_params() (combo.gp.core.model.model method),

15decomp_params() (combo.gp.core.prior method), 16decomp_params() (combo.gp.cov.gauss.gauss method),

16delete() (combo.variable.variable method), 11delete_actions() (combo.search.discrete.policy.policy

method), 14delete_stats() (combo.blm.predictor.predictor method),

18delete_stats() (combo.gp.predictor.predictor method), 15delete_stats() (combo.predictor.base_predictor method),

11delete_t() (combo.variable.variable method), 11delete_X() (combo.variable.variable method), 11delete_Z() (combo.variable.variable method), 11disp_marlik() (combo.gp.core.learning.online method),

15

Eeval_marlik() (combo.gp.core.model.model method), 15export_all_sequence_best_fx()

(combo.search.discrete.results.history method),14

export_blm() (combo.gp.core.model.model method), 15export_history() (combo.search.discrete.policy.policy

method), 14export_predictor() (combo.search.discrete.policy.policy

method), 14export_sequence_best_fx()

(combo.search.discrete.results.history method),14

export_training() (combo.search.discrete.policy.policymethod), 14

Ffit() (combo.blm.predictor.predictor method), 18fit() (combo.gp.core.model.model method), 16fit() (combo.gp.predictor.predictor method), 15fit() (combo.predictor.base_predictor method), 11

Gget_actions() (combo.search.discrete.policy.policy

method), 14get_basis() (combo.blm.basis.fourier method), 18get_basis() (combo.blm.lik.gauss.gauss method), 20get_basis() (combo.blm.predictor.predictor method), 18get_basis() (combo.gp.predictor.predictor method), 15get_basis() (combo.predictor.base_predictor method), 11get_cand_params() (combo.gp.core.model.model

method), 16get_cand_params() (combo.gp.cov.gauss.gauss method),

17get_cand_params() (combo.gp.lik.gauss.gauss method),

18get_cand_params() (combo.gp.mean.const.const

method), 18get_cov() (combo.blm.lik.gauss.gauss method), 20get_cov() (combo.blm.prior.gauss.cov_const method), 20get_cov() (combo.blm.prior.gauss.gauss method), 21get_cov() (combo.gp.core.prior method), 16get_cov() (combo.gp.cov.gauss.gauss method), 16get_cov() (combo.gp.lik.gauss.gauss method), 18get_grad() (combo.gp.cov.gauss.gauss method), 16get_grad() (combo.gp.lik.gauss.gauss method), 18get_grad() (combo.gp.mean.const.const method), 18get_grad_cov() (combo.gp.core.prior method), 16get_grad_marlik() (combo.gp.core.model.model

method), 16get_grad_mean() (combo.gp.core.prior method), 16get_marginal_score() (combo.search.discrete.policy.policy

method), 14get_mean() (combo.blm.lik.gauss.gauss method), 20get_mean() (combo.blm.lik.linear.linear method), 20get_mean() (combo.blm.prior.gauss.gauss method), 21get_mean() (combo.gp.core.prior method), 16get_mean() (combo.gp.mean.const.const method), 18get_one_update() (combo.gp.core.learning.adam

method), 15get_one_update() (combo.gp.core.learning.online

method), 15get_params_bound() (combo.gp.core.model.model

method), 16get_params_bound() (combo.gp.cov.gauss.gauss

method), 16get_params_bound() (combo.gp.lik.gauss.gauss method),

18get_params_bound() (combo.gp.mean.const.const

method), 18

30 Index


get_post_fcov() (combo.blm.core.model.model method),19

get_post_fcov() (combo.blm.predictor.predictor method),18

get_post_fcov() (combo.gp.core.model.model method),16

get_post_fcov() (combo.gp.predictor.predictor method),15

get_post_fcov() (combo.predictor.base_predictormethod), 11

get_post_fmean() (combo.blm.core.model.modelmethod), 19

get_post_fmean() (combo.blm.predictor.predictormethod), 18

get_post_fmean() (combo.gp.core.model.model method),16

get_post_fmean() (combo.gp.predictor.predictor method),15

get_post_fmean() (combo.predictor.base_predictormethod), 11

get_post_params() (combo.blm.predictor.predictormethod), 18

get_post_params() (combo.gp.predictor.predictormethod), 15

get_post_params() (combo.predictor.base_predictormethod), 12

get_post_params_mean() (combo.blm.core.model.modelmethod), 19

get_post_samples() (combo.blm.predictor.predictormethod), 18

get_post_samples() (combo.gp.predictor.predictormethod), 15

get_post_samples() (combo.predictor.base_predictormethod), 12

get_prec() (combo.blm.lik.gauss.gauss method), 20get_prec() (combo.blm.prior.gauss.cov_const method),

20get_prec() (combo.blm.prior.gauss.gauss method), 21get_predict_samples() (combo.blm.predictor.predictor

method), 18get_predict_samples() (combo.gp.predictor.predictor

method), 15get_random_action() (combo.search.discrete.policy.policy

method), 14get_score() (combo.search.discrete.policy.policy

method), 14get_subset() (combo.variable.variable method), 11

Iinit_params() (combo.gp.mean.const.const method), 18init_params_search() (combo.gp.core.learning.batch

method), 15init_params_search() (combo.gp.core.learning.online

method), 15

Lload() (combo.gp.cov.gauss.gauss method), 16load() (combo.misc.set_config.adam method), 12load() (combo.misc.set_config.batch method), 12load() (combo.misc.set_config.learning method), 12load() (combo.misc.set_config.online method), 12load() (combo.misc.set_config.search method), 12load() (combo.misc.set_config.set_config method), 12load() (combo.search.discrete.policy.policy method), 14load() (combo.search.discrete.results.history method), 14load() (combo.variable.variable method), 11

Mmodel (combo.predictor.base_predictor attribute), 11

Oone_run() (combo.gp.core.learning.batch method), 15one_run() (combo.gp.core.learning.online method), 15

Ppost_sampling() (combo.blm.core.model.model method),

19post_sampling() (combo.gp.core.model.model method),

16predict_sampling() (combo.blm.core.model.model

method), 19predict_sampling() (combo.gp.core.model.model

method), 16prepare() (combo.blm.core.model.model method), 19prepare() (combo.blm.predictor.predictor method), 18prepare() (combo.gp.core.model.model method), 16prepare() (combo.gp.cov.gauss.gauss method), 16prepare() (combo.gp.predictor.predictor method), 15prepare() (combo.predictor.base_predictor method), 11print_params() (combo.gp.core.model.model method), 16print_params() (combo.gp.cov.gauss.gauss method), 16

Rrand_expans() (combo.gp.cov.gauss.gauss method), 16random_search() (combo.search.discrete.policy.policy

method), 14rest() (combo.gp.core.learning.adam method), 15run() (combo.gp.core.learning.batch method), 15run() (combo.gp.core.learning.online method), 15run() (combo.opt.adam method), 15

Ssampling() (combo.blm.core.model.model method), 19sampling() (combo.blm.lik.gauss.gauss method), 20sampling() (combo.gp.core.prior method), 16sampling() (combo.gp.lik.gauss.gauss method), 18save() (combo.gp.cov.gauss.gauss method), 16save() (combo.search.discrete.results.history method), 14

Index 31


save() (combo.variable.variable method), 11set_bias() (combo.blm.lik.gauss.gauss method), 20set_bias() (combo.blm.lik.linear.linear method), 20set_cov_params() (combo.gp.core.prior method), 16set_mean_params() (combo.gp.core.prior method), 16set_params() (combo.blm.basis.fourier method), 18set_params() (combo.blm.lik.gauss.gauss method), 20set_params() (combo.blm.lik.linear.linear method), 20set_params() (combo.blm.prior.gauss.cov_const method),

20set_params() (combo.blm.prior.gauss.gauss method), 21set_params() (combo.gp.core.model.model method), 15set_params() (combo.gp.core.prior method), 16set_params() (combo.gp.cov.gauss.gauss method), 16set_params() (combo.gp.lik.gauss.gauss method), 18set_params() (combo.gp.mean.const.const method), 18set_params() (combo.opt.adam method), 14set_seed() (combo.search.discrete.policy.policy method),

14show() (combo.blm.basis.fourier method), 18show() (combo.misc.set_config.adam method), 12show() (combo.misc.set_config.batch method), 12show() (combo.misc.set_config.learning method), 12show() (combo.misc.set_config.online method), 12show() (combo.misc.set_config.search method), 12show() (combo.misc.set_config.set_config method), 12sub_sampling() (combo.gp.core.model.model method),

15supp_params() (combo.gp.cov.gauss.gauss method), 16supp_params() (combo.gp.lik.gauss.gauss method), 18supp_params() (combo.gp.mean.const.const method), 18

Tt (combo.variable.variable attribute), 11trans_params() (combo.gp.lik.gauss.gauss method), 18

Uupdate() (combo.blm.predictor.predictor method), 18update() (combo.opt.adam method), 15update_stats() (combo.blm.core.model.model method),

19

Wwrite() (combo.search.discrete.policy.policy method), 14write() (combo.search.discrete.results.history method),

14

XX (combo.variable.variable attribute), 11

ZZ (combo.variable.variable attribute), 11

32 Index

combo documentation - read the docs · return type return score combo.search.score.pi(predictor,...

Documents