Approaches¶
HyperOpt¶
- publications:
[BYC13].
- bibtex:
bergstra2013making- repository:
- supports:
cont,int,cat- approach-key:
hyperopt- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 3)
- --int-conversion-mode
How to convert integer variables. Choose from
quniformorrandint. (default:quniform)
Random Search (via HyperOpt)¶
SMAC3¶
- publications:
- bibtex:
hutter2010sequential-extended,hutter2011sequential- repository:
- supports:
cont,int,cat- approach-key:
smac- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 3)
- --deterministic
Whether to run SMAC in deterministic mode. (default: false)
Warning
Non-deterministic mode spends a significant portion of its evaluation budget re-evaluating previous solutions. Enabling deterministic mode disables this, and may therefore provide better performance.
DONE¶
- publications:
- bibtex:
DONEpaper- repository:
- supports:
cont- approach-key:
donejl- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 0)
- --n-basis
The number of basis functions (default: 2000)
- --sigma-coeff
Variance of initial random fourier expansion coefficients. (default: \(\min(0.1, {d}^{-0.5})\), with \(d\) the dimensionality of the function under test)
- --sigma-s
Variance for surrogate exploration (default: \(\min(0.1, {d}^{-0.5})\), with \(d\) the dimensionality of the function under test)
- --sigma-f
Variance for function exploration (default: \(\min(0.1, {d}^{-0.5})\), with \(d\) the dimensionality of the function under test)
Note
DONE is sensitive to its parameters. The default sigma values are for a normalized search space. If either the input or output values are very large, performance may suffer.
IDONE¶
- publications:
[BVdW19]
- bibtex:
bliek2019black- repository:
- supports:
int,cat(as interpreted as integer / binarized)- approach-key:
idone- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 5)
- --model
The kind of model to use. Choose between
basicandadvanced. (default:advanced)- --binarize-categorical
Whether to binarize categorical variables. Will turn a categorical variable with \(k\) possible values, into \(\log_2(k)\) binary (0 or 1) categorical variables. (default: false)
- --binarize-int
Whether to binarize integer variables. Similar to
--binarize-categorical, will turn a integer variable with \(k\) possible values, into \(\log_2(k)\) binary (0 or 1) integer variables. (default: false)- --sampling
What kind of random sampling to perform to motivate exploration. Can be
none,thompsonoruniform. (default:none)- --scaling
Whether to perform scaling based on the first sample. (default: false)
- --expl-prob
Sets the probability of performing an exploration step for each variable. Can be
normalorlarger. (default:normal)- --internal-logging
Whether to emit
IDONE’s internal logfiles. (default: false)
MVRSM¶
- publications:
[BVdW20]
- bibtex:
bliek2020black- repository:
- supports:
cont,int,cat(as interpreted as integer / binarized).Note
Note that behaviour differs (defaults to a fixed 1000 of basis functions) in the case that the function is only continuous.
- approach-key:
mvrsm- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 5)
- --model
The kind of model to use. Choose between
basicandadvanced. (default:advanced)- --binarize-categorical
Whether to binarize categorical variables. Will turn a categorical variable with \(k\) possible values, into \(\log_2(k)\) binary (0 or 1) categorical variables. (default: false)
- --scaling
Whether to perform scaling based on the first sample. (default: false)
CoCaBO¶
- publications:
[RAN+19]
- bibtex:
ru2019bayesian- repository:
- supports:
cont,int(interpreted as categorical),cat.Currently requires at least one continuous (
cont) and one discrete (int,cat) variable.- approach-key:
cocabo- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 24)
bayesianoptimization¶
pyGPGO¶
- publications:
[JG17]
- bibtex:
pygpgo,Jimenez2017- repository:
- supports:
cont,int(via rounding),cat(interpreted as integer, via rounding)Note
Built-in support for integers is not used due to crashes.
- approach-key:
pygpgo- parameters:
- --rand-evals
The number of random evaluations to perform before utilizing the surrogate model. (default: 3)