jumping the gun on tasks with dependencies?

[Nathan Schneider]

Circumstances: train_nodict/Baseline.baseline is running in the scheduler. I added some test conditions and started a new instance of ducttape to add them to the queue. Surprisingly, it tried to run them immediately even though train_nodict hadn't yet finished:

Acquiring lock for train_nodict/Baseline.baseline
Acquiring lock for test_nodict/WhichIters.200+WhichMaxIters.200
Acquiring lock for test_nodict/WhichIters.100+WhichMaxIters.200
Acquiring lock for test_nodict/DevOrTest.test+WhichIters.200+WhichMaxIters.200
Acquiring lock for test_nodict/DevOrTest.test+WhichAnn.annB+WhichIters.200+WhichMaxIters.200
Running test_nodict/DevOrTest.test+WhichAnn.annB+WhichIters.200+WhichMaxIters.200 in /share/mal2/nschneid/locally-normalized-EM-sequence-model/./test_nodict/DevOrTest.test+WhichAnn.annB+WhichIters.200+WhichMaxIters.200
Using submitter pbs
Running test_nodict/DevOrTest.test+WhichIters.200+WhichMaxIters.200 in /share/mal2/nschneid/locally-normalized-EM-sequence-model/./test_nodict/DevOrTest.test+WhichIters.200+WhichMaxIters.200

(...test_nodict subsequently fails due to a bug in the task body.)

The relevant task headers:

task train_nodict < feats=/mal2/nschneid/locally-normalized-EM-sequence-model/ar-data/wiki.unlabeled.bio.head100k3.cut-f1-26,31,34.stemfeats > sstmodel :: T=51 MAXITERS=(WhichMaxIters: 5000 200) PRINTRATE=(WhichMaxIters: 5000=100 200=50) .submitter=pbs .walltime="24:00:00" .cpus=1 .mem=15g

task test_nodict < model=$sstmodel@train_nodict feats=$anns > pred :: iters=(WhichIters: 4000 2000 1000 800 600 400 200 100 5000 50 150 250) .submitter=pbs .walltime="05:00:00" .cpus=1 .mem=8g

and globals:

anns=(DevOrTest: 
      dev=(WhichAnn: annA=/mal2/nschneid/locally-normalized-EM-sequence-model/ar-data/wiki.dev.sst.annA.feats.sym.cut-f1-26,31,34,37.stemfeats  annB=/mal2/nschneid/locally-normalized-EM-sequence-model/ar-data/wiki.dev.sst.annB.feats.sym.cut-f1-26,31,34,37.stemfeats)
      test=(WhichAnn: annA=/mal2/nschneid/locally-normalized-EM-sequence-model/ar-data/wiki.ssttest.sst.annA.feats.sym.cut-f1-26,31,34,37.stemfeats  annB=/mal2/nschneid/locally-normalized-EM-sequence-model/ar-data/wiki.ssttest.sst.annB.feats.sym.cut-f1-26,31,34,37.stemfeats))

and plans:

reach train_nodict, test_nodict, eval via (WhichMaxIters: 5000) * (WhichIters: 100 200 400 1000 4000) * (WhichAnn: *)

reach train_nodict, test_nodict, eval via (WhichMaxIters: 200) * (WhichIters: 50 100 150 200) * (WhichAnn: *) * (DevOrTest: *)

Shouldn't the @train_nodict annotation have marked a dependency between the two tasks? If so, why is it trying to execute the second one before the first one is finished?

This is my first time using nested branch points, so am I misunderstanding something here?

Thanks,
Nathan

[Nathan Schneider]

Oh never mind, I forgot there were two realizations of train_nodict
due to WhichMaxIters, and one of them had already finished. :P