README.md

In PhilippScheller/VisBayesOpt: Visualizing model based optimization runs from package mlrMBO

VisBayesOpt

Visualization of Bayesian Optimization runs from package mlrMBO

General

All functions are implemented in R6 class design and require only 1 Object as input, a final.opt.state (returned by mbo()) from mlrMBO. VisBayesOptis build on mlr2 (since mlr3 is still under progress).

Installation

devtools::install_github("PhilippScheller/VisBayesOpt")

Documentation

All functions come with a documentation which can be accessed in the help page, e.g. for MboPlotProgress we can inspect the help page by

?MboPlotProgress

Visualization via Shiny App

You can deploy the app on your local machine via the following function

runAppLocal()

Workflow

We use the exemplary final.opt.stateobject mboObj2 which contains a mixed search space with numeric and discrete search space parameters. We show the worklfow for the R6 class MboPlotSearchSpace(). Due to the object oriented style of R6, the user can easily apply the workflow to any other class from VisBayesOpt.

Step 1 - Create a R6 object by calling the constructor with $new()

r{} library(VisBayesOpt) mboObj2 = readRDS("inst/test-data/mboObj2.rds") #adjust to local file mbo_plot_class = MboPlotSearchSpace$new(mboObj2)

Step 2 - Plot search space by calling the $plot() function on the R6 object

At this step we can already create our first plot with the default param_vals. Every R6 class (MboPlotProgress, MboPlotInputSpace,...) comes with the unified $plot() function, i.e. we can call $plot() on every R6 object from VisBayesOpt.

mbo_plot_class$plot()

Step 3 - Modify param_vals

All plot classes are inherits from the overall class MboPlot which contains a setting method we can use in all classes. Once the constructor is called (in Step 1) the param_valsof the plot are initialized with default values. We can change these values by using the $set_param_vals() function. $set_param_vals()requires a list input with all parameters to be specified (e.g. if you want to set only 1 param value you need to pass the other values with their default values; this is no problems since the functions only come with 1 or 2 param vals).

Now we do not want to include the points from the initial design so we set include_init_design = FALSE (to see the default values for all params of a class see the help page of the class), so we specify it by the function $set_param_vals() on the createt mbo_plot_class object:

mbo_plot_class$set_param_vals(list(include_y = TRUE, include_init_design = FALSE))

By doing so we tell the object mbo_plot_class to change the parameters of the function $plot().

Afterwards we can simply create the new plot by calling the $plot() function again and see that the changes are applied.

mbo_plot_class$plot()

PhilippScheller/VisBayesOpt documentation built on Sept. 14, 2020, 12:47 p.m.