Hi Robert!

Ok, good to know. Thanks!

Cheers!
Max

On Thu, Jan 8, 2015 at 4:16 PM, Robert Metzger <rmetzger@apache.org> wrote:
Hi Max,

No. I think there is nobody in the Flink community who has plans to implement nested iterations in the near future.

On Wed, Jan 7, 2015 at 10:58 AM, Maximilian Alber <alber.maximilian@gmail.com> wrote:
Thanks!

I made a workaround using a pseudo join with the workset. But now I'm back to the nested iteration issue. Is there any chance that this feature will be available in the next time(2-3 weeks)?

Cheers,
Max

On Wed, Jan 7, 2015 at 10:11 AM, Aljoscha Krettek <aljoscha@apache.org> wrote:
Hi,
the problem is that your operations do not depend on the iteration-step-dataset. Your code could be rewritten like this to make it more obvious:

 val emptyDataSet = env.fromCollection[Vector](Seq())
 // here we call the function
     val center = calcCenter(env, X, residual, randoms, -1)

     val centerX = (X subtV center) map {_ square}
     val x = calcWidthHeight(env, centerX, residual, widthCandidates, center)
     val width = x._1
     val height = x._2

     residual = residual - (getKernelVector(X, center, width) multV height)

     val centerOut = center map {x => new Vector(0, x.values)}
     val widthOut = width map {x => new Vector(1, x.values)}
     val heightOut = height map {x => new Vector(2, x.values)}
     val stepModel = centerOut union widthOut union height

   // here the loop begins
   val model = emptyDataSet.iterate(config.iterations){
     stepSet =>

     stepSet union (stepModel.map(new RichMapFunction[Vector, Vector]{
     def map(x: Vector) = new Vector((getIterationRuntimeContext.getSuperstepNumber-1)*3+x.id, x.values)
      }))
    }

    model map { _ toString } writeAsText config.outFile
}

This means that for the system these operations are considered to be outside the loop, thus you don't have access to the IterationContext.

Regards,
Aljoscha

On Tue, Jan 6, 2015 at 4:13 PM, Maximilian Alber <alber.maximilian@gmail.com> wrote:
Hey Flinksters!

ran into this error

java.lang.IllegalStateException: This stub is not part of an iteration step function.

below is my code, the concerning parts are marked. Is it a problem, that the stub is in a function that is called from the iteration step function?


Code:

    ......

   val emptyDataSet = env.fromCollection[Vector](Seq())
   // here the loop begins
   val model = emptyDataSet.iterate(config.iterations){
     stepSet =>
     // here we call the function
     val center = calcCenter(env, X, residual, randoms, -1)

     val centerX = (X subtV center) map {_ square}
     val x = calcWidthHeight(env, centerX, residual, widthCandidates, center)
     val width = x._1
     val height = x._2

     residual = residual - (getKernelVector(X, center, width) multV height)

     val centerOut = center map {x => new Vector(0, x.values)}
     val widthOut = width map {x => new Vector(1, x.values)}
     val heightOut = height map {x => new Vector(2, x.values)}
     val stepModel = centerOut union widthOut union height

     stepSet union (stepModel.map(new RichMapFunction[Vector, Vector]{
     def map(x: Vector) = new Vector((getIterationRuntimeContext.getSuperstepNumber-1)*3+x.id, x.values)
      }))
    }

    model map { _ toString } writeAsText config.outFile
}


def calcCenter(env: ExecutionEnvironment, X: DataSet[Vector], residual: DataSet[Vector], randoms: DataSet[Vector], iteration: Int): DataSet[Vector] = {
    val residual_2 = residual * residual
     val randomValue = if(iteration >= 0)
        (randoms filter {_.id == iteration})
     else
       // and this filter function causes the error
       (randoms.filter(new RichFilterFunction[Vector]{
           def filter(x: Vector) = x.id == (getIterationRuntimeContext.getSuperstepNumber-1)
     }))
    val ys = ((residual_2 sumV() neutralize) * (randomValue neutralize))

   .....

The full errror:

Error: The program execution failed: java.lang.IllegalStateException: This stub is not part of an iteration step function.
at org.apache.flink.api.common.functions.AbstractRichFunction.getIterationRuntimeContext(AbstractRichFunction.java:59)
at bumpboost.BumpBoost$$anon$2.filter(BumpBoost.scala:119)
at bumpboost.BumpBoost$$anon$2.filter(BumpBoost.scala:118)
at org.apache.flink.api.java.operators.translation.PlanFilterOperator$FlatMapFilter.flatMap(PlanFilterOperator.java:47)
at org.apache.flink.runtime.operators.chaining.ChainedFlatMapDriver.collect(ChainedFlatMapDriver.java:79)
at org.apache.flink.runtime.operators.chaining.ChainedMapDriver.collect(ChainedMapDriver.java:78)
at org.apache.flink.runtime.operators.DataSourceTask.invoke(DataSourceTask.java:176)
at org.apache.flink.runtime.execution.RuntimeEnvironment.run(RuntimeEnvironment.java:257)
at java.lang.Thread.run(Thread.java:745)


I append my program with the input files. To reproduce the error use following command line args, please replace in_file, random_file, width_candidates with the provided ones, and put for out_file the path you want to:

flink run -v bump_boost-0.1.jar -c bumpboost.Job in_file=foobar out_file=/tmp/tmphIHeEs random_file=/tmp/tmpeKQPZk dimensions=1 N=100 width_candidates_file=/tmp/tmpTJLKMm N_width_candidates=50 iterations=30 multi_bump_boost=0 gradient_descent_iterations=30 cache=False start_width=1.0 min_width=-4 max_width=6 min_width_update=1e-08 max_width_update=10

Thanks!
Cheers,
Max