[mlpack-svn] [MLPACK] #300: allknn fails for mnist8m dataset
MLPACK Trac
trac at coffeetalk-1.cc.gatech.edu
Wed Aug 14 18:47:07 EDT 2013
#300: allknn fails for mnist8m dataset
---------------------+------------------------------------------------------
Reporter: rozyang | Owner:
Type: defect | Status: new
Priority: major | Milestone:
Component: mlpack | Keywords:
Blocking: | Blocked By:
---------------------+------------------------------------------------------
I tried to build KNN for the mnist8m dataset (a larger version of MNIST,
with 8 million samples). The dataset is available at
http://ml.nec-labs.com/download/data/mnist8m/
% allknn -r mnist8m.csv -n mnist8m_3nn.csv -d distances_out.csv -k 3 -v
[INFO ] Loading 'mnist8m.csv' as CSV data. Segmentation fault
The same procedure works for MNIST with 70000 samples. I am wondering
whether there is some size limit in allknn. Or is it simply constrained by
memory? Currently I use 16G RAM. I am using mlpack version 1.0.5.
--
Ticket URL: <http://trac.research.cc.gatech.edu/fastlab/ticket/300>
MLPACK <www.fast-lab.org>
MLPACK is an intuitive, fast, and scalable C++ machine learning library developed at Georgia Tech.
More information about the mlpack-svn
mailing list