Errors by passing the design in mbo about mlrmbo HOT 22 CLOSED

This code does not cause any errors anymore. Line 57 now checks the column names of design.x.
Can we close this?

from mlrmbo.

But design.x was even wrong. I will try to check your changings soon. I also got erros caused in lines 70 und 71.

from mlrmbo.

I did not change anything :-) Your example code works perfectly with the generateMBODesign fun from the repository.

from mlrmbo.

Very strenge... I will look at the code and wirte some more comments.

from mlrmbo.

I still have problems with my application. if I run my example I get the message "Computing y column for design. Was not provided" which is not true.

ps in my example is wronng. The right one is:
ps = makeNumericParamSet(len = number_of_parameters(objfun), lower = lower_bounds(objfun), upper = upper_bounds(objfun))

And I still think that the problem is in line 57 (disign.x is false in this place)

from mlrmbo.

1. For source code please use the correct markup on github:

https://help.github.com/articles/github-flavored-markdown
see Fenced code blocks

2. Can you - just to make sure - paste a complete example that reproduces this error?

Actually, if you have that, you could yourself create a unit test. At an appropriate place.
That would be best!

from mlrmbo.

I created an unit test: initdesign_test.R.

As can be seen, although I provided y-vector for initial design, mbo “does not see it” und evaluates the initial design again.

from mlrmbo.

On my system I get a pretty satisfactory result. All your tests pass successfully. I cannot reproduce your problem.

library(methods)
library(testthat)
library(devtools)
library(BBmisc)
library(mlr)
Lade nötiges Paket: ParamHelpers
load_all(".", reset=TRUE)
test_file("inst/tests/test_mbo_initial_design.R")
# mbo initial design : ...

I guess I tested the correct unit test. You probably renamed the test file to test_mbo_initial_design.R?

from mlrmbo.

Oh, git blame told me, that you are not the author of test_mbo_initial_design.R. Could you please commit your test to let me/us comprehend your problem?

from mlrmbo.

@nbauer2: Could you please check in you unit test. Otherwise we cannot work on it. Moreover stick to the naming conventions of the tests, i. e., name your test test_myTestName.R.

from mlrmbo.

I pushed test_initdesign.R. I get a new error, which I have not got before (after mbo() command):

m=mbo(makeMBOFunction(objfun), design=design, par.set=ps, learner=learner_km, control=ctrl, show.info=TRUE)
Error in checkStuff(fun, par.set, design, learner, control) : 
  could not find function "hasDiscrete"

I updated mlr, mlrMOBO, ParamHelpers and BBmisc, but this function still could not be find.

from mlrmbo.

I guess you installed the current CRAN version of ParamHelpers, i.e., version 1.1-36. There are some new helper functions including hasDiscrete in the upcoming release. You should install the github developer version!

from mlrmbo.

I will piblish the new paramhelpers to cran this week.

Nadja these things we mention in the commit messages and news File. If you dont want to wait you can look there!

from mlrmbo.

And nadja , if appropriate, please push a regular unit test. Look at the ones already there.

You pushed something else.

It takes too much time to clean this up.

from mlrmbo.

Very strenge. I use Githup Version of ParamHelpers. I first pulled changes and then did make install. I will check now why the old version was loaded in my R. Sometimes it happens.

from mlrmbo.

@nbauer2: I converted your test to a unit test and moved it to tests/testthat and modified it a bit, because it was to time consuming.

from mlrmbo.

Very strenge. I use Githup Version of ParamHelpers. I first pulled changes and then did make
install. I will check now why the old version was loaded in my R. Sometimes it happens.

The only reason I can guess is if you save and load R workspaces in your IDE. Which you should never do IMHO because of exactly the problem you mentioned.

from mlrmbo.

Yes, I had to directories with R packages, now I deleted one, so that this error should not appear more.
@jakob: Thak you for converting.

This unit test produces error at the moment. And - more important - mbo does not see the initial design (first message: Computing y column for design. Was not provided )

from mlrmbo.

At the moment all tests fail due to a change of Daniel in evalTargetFunction in his last commit! Replace

z = parallelMap(fun2, xs)
z = parallelMap(fun2, xs, impute.error =
     function(x)
       list(y = rep(NA, control$number.of.targets), time = NA)
)

with z = parallelMap(fun2, xs)in evalTargetFunction and the test should run fine.

from mlrmbo.

Yes, there is no error message more. But my problem is that mbo evaluates the initial design once again (as if design=NULL) although it was provided!

from mlrmbo.

Please update parallelMap to the newest github version. The tests fail, 'cause i'm using a new feature of parallelMap.

from mlrmbo.

Yes, there is no error message more. But my problem is that mbo evaluates the initial design once
again (as if design=NULL) although it was provided!

@nbauer2: It works now. There was a bug in generateMBODesign.
Closing this issue.

from mlrmbo.