search

sancha / jrae

I re-implemented a semi-supervised recursive autoencoder in java. I think it is a pretty nice technique. Check it out! Or fork it
http://www.socher.org/index.php/Main/Semi-SupervisedRecursiveAutoencodersForPredictingSentimentDistributions
72 stars 41 forks source link

Converging Problem with FullRun #10

Closed pnvphuong closed 10 years ago

pnvphuong commented 10 years ago

Hi sancha, I am using RC3 and have good results with the RAEBuilder class. However, when I tried the FullRun class (my project needs to run RAE using K-fold cross validation), I encountered the converging problem. In particular, I have set -MaxIterations 70 . However, the program could not finish the first fold and said QNMinimizer terminated without converging FileNotFoundException: data/mov/data/mov/tunedTheta.rae.0.2.0.5.rae

I also attached the screenshot from my machine Do you have any solutions for this situation? Best, Phuong screenshot

sancha commented 10 years ago

That's interesting, as the minimizer suggests, please increase the number of iterations available for the minimizer to remove the error. Even if the minimizer terminates at that step, the solution should be reasonable, the error just says that there is scope to reduce the function further. The error is actually due to a misconfiguration during runtime. It did not find the directory /data/mov/data/mov/ . It should only be looking for /data/mov Could you check you config?

pnvphuong commented 10 years ago

Hi sancha, In FullRun.java line 76: tunedTheta.Dump(params.dir + "/" + params.ModelFile + params.AlphaCat + "." + params.Beta + ".rae");

I think we do not need to add params.dir + "/" because params.ModelFile already contains the full path (as suggested in the run.sh). That's why there is a duplication error as you have pointed out