Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Accessing the ensembled vectors #43

Closed
meghmehta opened this issue Feb 22, 2017 · 14 comments
Closed

Accessing the ensembled vectors #43

meghmehta opened this issue Feb 22, 2017 · 14 comments

Comments

@meghmehta
Copy link

after running "ninja", where can I find the ensemble vectors?

@rspeer
Copy link
Member

rspeer commented Feb 22, 2017

In this build process, the labels will appear in build-data/combo840.standardized.conceptnet5.labels and the vectors in build-data/combo840.l1.standardized.conceptnet5.npy. The filenames are elaborate because this repository builds many different variations of the system to compare in an evaluation.

There is also a new build process that's part of the conceptnet5 repository, which will give you more up-to-date vectors in data/vectors/numberbatch.h5, as well as the tools you need to look up ConceptNet nodes in that vector space.

@meghmehta
Copy link
Author

Thank you - my problem is that I can't find any build-data folder in the repository and so I am not sure where to find these files... would you be able to guid me in this?

@rspeer
Copy link
Member

rspeer commented Feb 23, 2017

That's where the files created by ninja go.

What happened when you ran ninja in the code/ directory? Did it crash?

@meghmehta
Copy link
Author

so I've been running the 16.04 branch....it basically runs through build_conceptnet_retrofitting() which is called in ninja.py. But there's only a source_data folder, I can't find a build_data folder...

@rspeer
Copy link
Member

rspeer commented Feb 23, 2017

Oh. You haven't run the build yet.

Running ninja.py outputs a file called build.ninja, which contains instructions to the Ninja build system (https://ninja-build.org/) for how to build ConceptNet Numberbatch. After that, you run ninja to actually run the build.

@meghmehta
Copy link
Author

Yes I did run ninja as well... like i ran ninja.py and then I ran ninja but I don't get any new outputs...

@meghmehta
Copy link
Author

I actually get the error:
Importerror: cannot import ninja from ninja

@rspeer
Copy link
Member

rspeer commented Feb 23, 2017

Can you copy and paste the actual output, please?

@rspeer
Copy link
Member

rspeer commented Feb 23, 2017

What's particularly peculiar about saying you got "Importerror: cannot import ninja from ninja" is that it looks almost like a Python error, and this makes me doubt you're actually running Ninja, which is written in C++.

@meghmehta
Copy link
Author

I actually don't have that error anymore...when I run ninja it gives the following output:
log: warning: no configuration file specified, using default values
log: ninja version 0.1.3 initializing
log: magic group: gid=0 (wheel)
log: entering main loop
log: generating initial pid array..
log: now monitoring process activity

How long would it take for it to complete? It's been running for > 1 hour now.

@rspeer
Copy link
Member

rspeer commented Feb 24, 2017

It should take a day or so. Keep in mind that this process runs all the conditions of the experiment.

@meghmehta
Copy link
Author

okay thank you!

@meghmehta
Copy link
Author

Is it GPU enabled?

@rspeer
Copy link
Member

rspeer commented Feb 24, 2017

Man, how difficult to set up do you want it to be?

This task is memory-constrained, so I think it would actually run much worse on a GPU.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants