在使用 H2O R model 进行预测时使用 optim package 时出错

Question

In R, there is a very convenient optimization package called "optim" . You can feed it a function, an initial position, and sometimes other control input, and it will optimize your function.

I am trying to use this with h2o.ai model in the following way:

make_model <- h2o.stuff(x,y,training, parameters)
f <- function(x,make_model){ h2o.predict() %>% ...}
f2 <- function(x){f(x,make_model)}

optim(start, function=f2, ...)

When I run test cases they work:

f2(start_point) 
f2(known_values)

These return exactly what I'm looking for. They return the values they should.

When I try to run the optim on the function, it doesn't want to work, and gives this error:

java.lang.IllegalArgumentException: Test/Validation dataset has no columns in common with the training set

java.lang.IllegalArgumentException: Test/Validation dataset has no columns in common with the training set
    at hex.Model.adaptTestForTrain(Model.java:1383)
    at hex.Model.adaptTestForTrain(Model.java:1222)
    at hex.Model.score(Model.java:1509)
    at water.api.ModelMetricsHandler$1.compute2(ModelMetricsHandler.java:396)
    at water.H2O$H2OCountedCompleter.compute(H2O.java:1557)
    at jsr166y.CountedCompleter.exec(CountedCompleter.java:468)
    at jsr166y.ForkJoinTask.doExec(ForkJoinTask.java:263)
    at jsr166y.ForkJoinPool$WorkQueue.runTask(ForkJoinPool.java:974)
    at jsr166y.ForkJoinPool.runWorker(ForkJoinPool.java:1477)
    at jsr166y.ForkJoinWorkerThread.run(ForkJoinWorkerThread.java:104)

Error: java.lang.IllegalArgumentException: Test/Validation dataset has no columns in common with the training set
Called from: doTryCatch(return(expr), name, parentenv, handler)

Now it looks to me like the C-based optim and the Java based h2o.ai don't want to play well together. The function is a bit nonlinear, but it evaluates the known points directly and properly.

Is there a decent way to get around this without changing architectures?

Answer 1

I don't think this error has anything to do with using the optim package. That is an error from the H2O package which just that "Test/Validation dataset has no columns in common with the training set" .

So you should check the column names of your test set (or whatever dataset you're passing to the h2o.predict() function) to make sure it looks like your training set (all training columns should be present in test set). If you can post a reproducible example, I can probably be more helpful.