Create a Feature Object

Create a FeatureObject, which will be used as input for all the feature computations.

createFeatureObject(init, X, y, fun, minimize, lower, upper, blocks, objective,
  force = FALSE)

Arguments

init	[data.frame] A data.frame, which can be used as initial design. If not provided, it will be created either based on the initial sample X and the objective values y or X and the function definition fun.
X	[data.frame or matrix] A data.frame or matrix containing the initial sample. If not provided, it will be extracted from init.
y	[numeric or integer] A vector containing the objective values of the initial design. If not provided, it will be extracted from init.
fun	[function] A function, which allows the computation of the objective values. If it is not provided, features that require additional function evaluations, can't be computed.
minimize	[logical(1)] Should the objective function be minimized? The default is TRUE.
lower	[numeric or integer] The lower limits per dimension.
upper	[numeric or integer] The upper limits per dimension.
blocks	[integer] The number of blocks per dimension.
objective	[character(1)] The name of the feature, which contains the objective values. The default is "y".
force	[logical(1)] Only change this parameter IF YOU KNOW WHAT YOU ARE DOING! Per default (force = FALSE), the function checks whether the total number of cells that you are trying to generate, is below the (hard-coded) internal maximum of 25,000 cells. If you set this parameter to TRUE, you agree that you want to exceed that internal limit. Note: *Exploratory Landscape Analysis (ELA)* is only useful when you are limited to a small budget (i.e., a small number of function evaluations) and in such scenarios, the number of cells should also be kept low!

Value

[FeatureObject].

Examples

# (1a) create a feature object using X and y:
X = createInitialSample(n.obs = 500, dim = 3,
  control = list(init_sample.lower = -10, init_sample.upper = 10))
y = apply(X, 1, function(x) sum(x^2))
feat.object1 = createFeatureObject(X = X, y = y,
  lower = -10, upper = 10, blocks = c(5, 10, 4))

# (1b) create a feature object using X and fun:
feat.object2 = createFeatureObject(X = X,
  fun = function(x) sum(sin(x) * x^2),
  lower = -10, upper = 10, blocks = c(5, 10, 4))

# (1c) create a feature object using a data.frame:
feat.object3 = createFeatureObject(iris[,-5], blocks = 5,
  objective = "Petal.Length")

# (2) have a look at the feature objects:
feat.object1
#> Feature Object:
#> - Number of Observations: 500
#> - Number of Variables: 3
#> - Lower Boundaries: -1.00e+01, -1.00e+01, -1.00e+01
#> - Upper Boundaries: 1.00e+01, 1.00e+01, 1.00e+01
#> - Name of Variables: x1, x2, x3
#> - Optimization Problem: minimize y
#> - Number of Cells per Dimension: 5, 10, 4
#> - Size of Cells per Dimension: 4.00, 2.00, 5.00
#> - Number of Cells:
#>   - total: 200
#>   - non-empty: 182 (91.00%)
#>   - empty: 18 (9.00%)
#> - Average Number of Observations per Cell:
#>   - total: 2.50
#>   - non-empty: 2.75
feat.object2
#> Feature Object:
#> - Number of Observations: 500
#> - Number of Variables: 3
#> - Lower Boundaries: -1.00e+01, -1.00e+01, -1.00e+01
#> - Upper Boundaries: 1.00e+01, 1.00e+01, 1.00e+01
#> - Name of Variables: x1, x2, x3
#> - Optimization Problem: minimize y
#> - Function to be Optimized: function (x) sum(sin(x) * x^2)
#> - Number of Cells per Dimension: 5, 10, 4
#> - Size of Cells per Dimension: 4.00, 2.00, 5.00
#> - Number of Cells:
#>   - total: 200
#>   - non-empty: 182 (91.00%)
#>   - empty: 18 (9.00%)
#> - Average Number of Observations per Cell:
#>   - total: 2.50
#>   - non-empty: 2.75
feat.object3
#> Feature Object:
#> - Number of Observations: 150
#> - Number of Variables: 3
#> - Lower Boundaries: 4.30e+00, 2.00e+00, 1.00e-01
#> - Upper Boundaries: 7.90e+00, 4.40e+00, 2.50e+00
#> - Name of Variables: Sepal.Length, Sepal.Width, Petal.Width
#> - Optimization Problem: minimize Petal.Length
#> - Number of Cells per Dimension: 5, 5, 5
#> - Size of Cells per Dimension: 0.72, 0.48, 0.48
#> - Number of Cells:
#>   - total: 125
#>   - non-empty: 37 (29.60%)
#>   - empty: 88 (70.40%)
#> - Average Number of Observations per Cell:
#>   - total: 1.20
#>   - non-empty: 4.05

# (3) now, one could calculate features
calculateFeatureSet(feat.object1, "ela_meta")
#> $ela_meta.lin_simple.adj_r2
#> [1] -0.005733815
#> 
#> $ela_meta.lin_simple.intercept
#> [1] 96.88311
#> 
#> $ela_meta.lin_simple.coef.min
#> [1] 0.1943708
#> 
#> $ela_meta.lin_simple.coef.max
#> [1] 0.3058714
#> 
#> $ela_meta.lin_simple.coef.max_by_min
#> [1] 1.573649
#> 
#> $ela_meta.lin_w_interact.adj_r2
#> [1] 0.006650439
#> 
#> $ela_meta.quad_simple.adj_r2
#> [1] 1
#> 
#> $ela_meta.quad_simple.cond
#> [1] 1
#> 
#> $ela_meta.quad_w_interact.adj_r2
#> [1] 1
#> 
#> $ela_meta.costs_fun_evals
#> [1] 0
#> 
#> $ela_meta.costs_runtime
#> [1] 0.017
#> 
calculateFeatureSet(feat.object2, "cm_grad")
#> $cm_grad.mean
#> [1] 0.655673
#> 
#> $cm_grad.sd
#> [1] 0.1791901
#> 
#> $cm_grad.costs_fun_evals
#> [1] 0
#> 
#> $cm_grad.costs_runtime
#> [1] 0.056
#> 
library(plyr)
calculateFeatureSet(feat.object3, "cm_angle", control = list(cm_angle.show_warnings = FALSE))
#> $cm_angle.dist_ctr2best.mean
#> [1] 1.886064
#> 
#> $cm_angle.dist_ctr2best.sd
#> [1] 0.7072052
#> 
#> $cm_angle.dist_ctr2worst.mean
#> [1] 1.909764
#> 
#> $cm_angle.dist_ctr2worst.sd
#> [1] 0.6994935
#> 
#> $cm_angle.angle.mean
#> [1] 6.260766
#> 
#> $cm_angle.angle.sd
#> [1] 7.44275
#> 
#> $cm_angle.y_ratio_best2worst.mean
#> [1] 0.0650481
#> 
#> $cm_angle.y_ratio_best2worst.sd
#> [1] 0.07084862
#> 
#> $cm_angle.costs_fun_evals
#> [1] 0
#> 
#> $cm_angle.costs_runtime
#> [1] 0.147
#>