Open minhlab opened 7 years ago
When I run it on real data, I got this error:
$ ./oxlm.job
################################
# Config Summary
# order = 5
# feature context size = 5
# word_width = 150
# model-out = output/lb-ec-chain.bin
# input = output/ec_chain_as_sentences.train.unk.txt
# minibatch size = 10000
# minibatch threshold = 20000
# lambda LBL = 2
# lambda maxent = 2
# step size = 0.06
# iterations = 20
# evaluate frequency = 1000
# threads = 8
# randomise = 1
# diagonal contexts = 1
# activation = 2
# noise samples = 10
# hash space = 0
################################
Binned 12597 types in 100 classes with an average of 125.97 types per bin.
Reading training corpus...
...................
Done reading training corpus...
Reading test corpus...
Done reading test corpus...
Done creating the n-gram filter...
Creating the n-gram filter took 0 seconds...
........
Done creating the feature matcher...
Creating the feature matcher took 27.883 seconds...
===============================
Model parameters:
Context vocab size = 12597
Output vocab size = 12597
Total parameters = 3807397
===============================
./oxlm.job: line 15: 28847 Floating point exception
Turns out that I need to specify hash-space (default is 0 which causes division error). A meaningful error message would be useful.
Hi, the same issue of failed tests. Ubuntu 16 LTS.
I downloaded and built OxLM today and two tests failed. What should I do now?
I was using CentOS Linux 7.