Experimenting with even easier Torch installation

988 views
Skip to first unread message

Alex W

unread,
Apr 23, 2015, 8:11:14 AM4/23/15
to tor...@googlegroups.com
Hello,
I've been packaging up LuaJIT and Torch packages for even easier install using Continuum.IO's "conda" tool (http://conda.pydata.org/).
The goal is to turn all of the LuaJIT numerical libraries into reliable, fast, architecture-independent, one-line installs.

Here's how you can, from scratch, get all the packages you need to train a basic neural net on the CPU.

First, install anaconda, if you don't have it (you can check by typing "conda help", and if nothing useful comes up, then try the lines below)

# Install Anaconda on Linux:
bash Miniconda-latest-Linux-x86_64.sh -b -p $HOME/anaconda

# Install Anaconda on OS X
bash Miniconda-latest-MacOSX-x86_64.sh -b -p $HOME/anaconda



# Then, to install all the packages you need for training a neural network on the CPU, type:
conda install -c alexbw lua-nn lua-trepl lua-optim # That's it!




I'm just at the beginning stages of this, so please let me know if there's hiccups, and I'll try to work through it. 

You can look at all the build recipes here:

And if you want to submit a pull request for a new package (please do!), here's a walk-through explanation of how these recipes work:

subu

unread,
Apr 23, 2015, 3:51:46 PM4/23/15
to tor...@googlegroups.com
Hi Alex,
I see that miniconda is available for windows as well. Does this mean torch on windows ?

Alex Wiltschko

unread,
Apr 23, 2015, 4:11:47 PM4/23/15
to torch7 on behalf of subu
I haven't done any work on Windows yet. I would be delighted to help anyone who would like to try, though.
--
You received this message because you are subscribed to a topic in the Google Groups "torch7" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/torch7/KlGdXu9Bkgc/unsubscribe.
To unsubscribe from this group and all its topics, send an email to torch7+un...@googlegroups.com.
To post to this group, send email to tor...@googlegroups.com.
Visit this group at http://groups.google.com/group/torch7.
For more options, visit https://groups.google.com/d/optout.

Alex W

unread,
Apr 3, 2016, 5:31:08 AM4/3/16
to torch7
I've been working on this very slowly, but it now works. Install with the following-command.

conda install lua=5.2 lua-science -c alexbw

Options for lua= are lua=2.0, lua=5.1, lua=5.2, lua=5.3.
OSX/Linux only. CUDA not built for Linux yet, but working on that. +1 the CUDA issue on the github repo to indicate desire.

A

tdp...@gmail.com

unread,
Jun 21, 2016, 9:00:46 PM6/21/16
to torch7
Hi A,

I am really interested in your work and would like to help too. Right now the two painful package installation is cutorch and cunn. 

Please let me know if there is any easy way to do it. 

Cheers, 
Zhong
Reply all
Reply to author
Forward
0 new messages