0

I'm creating a design consisting of a set of tasks, imputation + learners (GraphLearner), and a CV-resampling. When I run all of it through 'benchmark' I get a R session crash (within RStudio) and I have to restart. I've tried it on 2 different PCs multiple times. Also memory and CPU usage remain moderate during the whole time.

This is the relevant line from rsession log file:

28 Apr 2021 09:01:17 [rsession-User] ERROR system error 10053 (An established connection was aborted by the software in your host machine) [request-uri: /events/get_events]; OCCURRED AT void __cdecl rstudio::session::HttpConnectionImpl::sendResponse(const class rstudio::core::http::Response &) src/cpp/session/http/SessionWin32HttpConnectionListener.cpp:113; LOGGED FROM: void __cdecl rstudio::session::HttpConnectionImpl::sendResponse(const class rstudio::core::http::Response &) src/cpp/session/http/SessionWin32HttpConnectionListener.cpp:118

Below is the 'design' that I created (it has 288 subtasks to be run):

> print(design)
               task            learner         resampling
  1: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
  2: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
  3: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
  4: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
  5: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
 ---                                                     
284: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
285: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
286: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
287: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>
288: <TaskRegr[43]> <GraphLearner[33]> <ResamplingCV[19]>

Looking at the error log above, it seems to be a connection issue. But, everything runs locally on my PC. I access the initial data file from a shared drive but once it's loaded in an object there are no more connections being made.

Has anyone seen this before?

Upon searching the web, I got a hint that it might be to do with parallelisation, but that was for an older version. Taking advice in that thread itself I updated all my packages again. I'm also using the latest versions of both R and RStudio (as of posting this).

intelinsight
  • 63
  • 5
  • I just re-ran 'benchmark(head(design))' which has just 6 sub-tasks but that also crashes R session. – intelinsight Apr 28 '21 at 09:22
  • Without a reprex and your environment, I suspect nobody might be able to help :) – pat-s Apr 28 '21 at 09:37
  • You could try [encapsulation](https://mlr3book.mlr-org.com/error-handling.html#encapsulation). This prevents that a crash of a single train or predict process causes the whole process to fail. – jakob-r Apr 28 '21 at 09:57

0 Answers0