Error: Error in as.data.frame.OptPathDF(opt.path, include.rest = FALSE) : No elements where selected (via 'dob' and 'eol')!
swaheera opened this issue · 8 comments
I recently got this error while trying to perform multiobjective constrained optimization with the mlrMBO library:
library(mlrMBO)
obj.fn = makeMultiObjectiveFunction(
name = "My test function",
fn = function(x1, x2, x3, x4) {
var_1 <- sin(x1 + x2)
var_2 <- cos(x1 - x2)
var_3 <- x1 + x4
var_4 <- x3 + x4 -7
goal_1 = sum(var_1 + var_2 + var_3 + var_4)
goal_2 = var_1 + var_2 - var_3 + var_4
goal_3 = var_1 + var_2 - var_3 + 2*var_4
return(c(goal_1, goal_2, goal_3))
},
n.objectives = 3L,
#define acceptable ranges
par.set = makeParamSet(
makeNumericParam("x1", lower = 20, upper = 40),
makeNumericParam("x2", lower = 30, upper = 45),
makeNumericParam("x3", lower = 10, upper = 20),
makeNumericParam("x4", lower = 10, upper = 50)
#define constraints
, forbidden = expression(x2 >x1 | x3 > x4)
),
minimize=rep(TRUE,3)
)
#create control gird
control=makeMBOControl(propose.points=1, final.method="best.predicted", final.evals=10)
control=setMBOControlTermination(control, iters=10)
control=setMBOControlInfill(control, crit=makeMBOInfillCritEI())
#perform optimization
lrn=makeMBOLearner(control, obj.fun)
res = mbo(obj.fun, design = NULL, learner = lrn, control = ctrl, show.info = TRUE)
#error
Warning in generateDesign(n.params * 4L, par.set, fun = lhs::maximinLHS) :
generateDesign could only produce 0 points instead of 16!
Computing y column(s) for design. Not provided.
Error in as.data.frame.OptPathDF(opt.path, include.rest = FALSE) :
No elements where selected (via 'dob' and 'eol')!
I am a bit confused. Where exactly do I need to specify "dob" and "eol"?
Thanks
Just by looking at the constraints it looks like they are too narrow (and partly don't make sense)
Forbidden: x2 >x1 | x3 > x4
In other words: x2 should always be smaller or equal then x1 AND x3 always has to be smaller then x4.
Having this in mind it does not make sense to have a higher lower bound for x2 then for x1.
Probably there is no feasible point found in the initial design.
I tried removing the constraints all together, but now I get a different error:
library(mlrMBO)
obj.fn = makeMultiObjectiveFunction(
name = "My test function",
fn = function(x1, x2, x3, x4) {
var_1 <- sin(x1 + x2)
var_2 <- cos(x1 - x2)
var_3 <- x1 + x4
var_4 <- x3 + x4 -7
goal_1 = sum(var_1 + var_2 + var_3 + var_4)
goal_2 = var_1 + var_2 - var_3 + var_4
goal_3 = var_1 + var_2 - var_3 + 2*var_4
return(c(goal_1, goal_2, goal_3))
},
n.objectives = 3L,
#define acceptable ranges
par.set = makeParamSet(
makeNumericParam("x1", lower = 20, upper = 40),
makeNumericParam("x2", lower = 30, upper = 45),
makeNumericParam("x3", lower = 10, upper = 20),
makeNumericParam("x4", lower = 10, upper = 50)
),
minimize=rep(TRUE,3)
)
#create control gird
control=makeMBOControl(propose.points=1, final.method="best.predicted", final.evals=10)
control=setMBOControlTermination(control, iters=10)
control=setMBOControlInfill(control, crit=makeMBOInfillCritEI())
#perform optimization
lrn=makeMBOLearner(control, obj.fun)
res = mbo(obj.fun, design = NULL, learner = lrn, control = ctrl, show.info = TRUE)
Error:
Error in checkClass(x, classes, ordered, null.ok) :
object 'obj.fun' not found
Is there any way to fix this?
Thanks
You named your function obj.fn and not obj.fun
Thank you for the correction - I found another typo : "ctrl" vs "control". I fixed both of these errors but now I have a new error.
library(mlrMBO)
obj.fn = makeMultiObjectiveFunction(
name = "My test function",
fn = function(x1, x2, x3, x4) {
var_1 <- sin(x1 + x2)
var_2 <- cos(x1 - x2)
var_3 <- x1 + x4
var_4 <- x3 + x4 -7
goal_1 = sum(var_1 + var_2 + var_3 + var_4)
goal_2 = var_1 + var_2 - var_3 + var_4
goal_3 = var_1 + var_2 - var_3 + 2*var_4
return(c(goal_1, goal_2, goal_3))
},
n.objectives = 3L,
#define acceptable ranges
par.set = makeParamSet(
makeNumericParam("x1", lower = 20, upper = 40),
makeNumericParam("x2", lower = 30, upper = 45),
makeNumericParam("x3", lower = 10, upper = 20),
makeNumericParam("x4", lower = 10, upper = 50)
#define constraints
, forbidden = expression(x2 >x1 | x3 > x4)
),
minimize=rep(TRUE,3)
)
#create control gird
control=makeMBOControl(propose.points=1, final.method="best.predicted", final.evals=10)
control=setMBOControlTermination(control, iters=10)
control=setMBOControlInfill(control, crit=makeMBOInfillCritEI())
#perform optimization
lrn=makeMBOLearner(control, obj.fn)
res = mbo(obj.fn, design = NULL, learner = lrn, control = control, show.info = TRUE)
I got the following error:
Warning in generateDesign(n.params * 4L, par.set, fun = lhs::maximinLHS) :
generateDesign could only produce 15 points instead of 16!
Error in checkStuff(fun, design, learner, control) :
Objective function has 3 objectives, but the control object assumes 1.
I tried to fix this by changing the number of objectives:
#create control gird
control=makeMBOControl(propose.points=1, final.method="best.predicted", n.objectives = 3L, final.evals=10)
control=setMBOControlTermination(control, iters=10)
control=setMBOControlInfill(control, crit=makeMBOInfillCritEI())
#perform optimization
lrn=makeMBOLearner(control, obj.fn)
res = mbo(obj.fn, design = NULL, learner = lrn, control = control, show.info = TRUE)
But then I got a new error:
Error in checkStuff(fun, design, learner, control) :
Setting of final.method and final.evals for multi-objective optimization not supported at the moment.
Do you have any idea why this error is being produced?
Thank you so much for your help!
Thanks
In MOO there is no final best but only a set of points that are pareto optimal.
You have to change makeMBOControl
control=makeMBOControl(propose.points=1)
Hello Dr. Richter,
Thank you for your reply.
I removed the "best.predicted" statement, but the code is still not working:
library(mlrMBO)
obj.fn = makeMultiObjectiveFunction(
name = "My test function",
fn = function(x1, x2, x3, x4) {
var_1 <- sin(x1 + x2)
var_2 <- cos(x1 - x2)
var_3 <- x1 + x4
var_4 <- x3 + x4 -7
goal_1 = sum(var_1 + var_2 + var_3 + var_4)
goal_2 = var_1 + var_2 - var_3 + var_4
goal_3 = var_1 + var_2 - var_3 + 2*var_4
return(c(goal_1, goal_2, goal_3))
},
n.objectives = 3L,
#define acceptable ranges
par.set = makeParamSet(
makeNumericParam("x1", lower = 20, upper = 40),
makeNumericParam("x2", lower = 30, upper = 45),
makeNumericParam("x3", lower = 10, upper = 20),
makeNumericParam("x4", lower = 10, upper = 50)
#define constraints
, forbidden = expression(x2 >x1 | x3 > x4)
),
minimize=rep(TRUE,3)
)
#create control gird
control=makeMBOControl(propose.points=1, n.objectives = 3L, final.evals=10)
#perform optimization
lrn=makeMBOLearner(control, obj.fn)
res = mbo(obj.fn, design = NULL, learner = lrn, control = control, show.info = TRUE)
But this produces the following error
Error in checkStuff(fun, design, learner, control) :
Setting of final.method and final.evals for multi-objective optimization not supported at the moment.
As a result of this error, I tried to remove "final.evals":
#create control gird
control=makeMBOControl(propose.points=1, n.objectives = 3L)
#perform optimization
lrn=makeMBOLearner(control, obj.fn)
res = mbo(obj.fn, design = NULL, learner = lrn, control = control, show.info = TRUE)
But now I get a different error (even though I have specified there are 3 objectives):
Warning in generateDesign(n.params * 4L, par.set, fun = lhs::maximinLHS) :
generateDesign could only produce 15 points instead of 16!
Error in checkStuff(fun, design, learner, control) :
Objective function has 3 objectives, but the control object assumes 1.
If you have time, can you please try running this code on your computer and see if you can get it to work? I have been trying to get this to work for a while, but without any results.
Your Help Is Greatly Appreciated,
Thanks
Hello Dr. Richter,,
Can you please take a look at this if you have some time?
Thanks