Closed ajratner closed 9 years ago
That's a great idea! Please check this in.
Sure, will do tomorrow
On Tue, Aug 4, 2015 at 12:29 AM Raphael Hoffmann notifications@github.com wrote:
That's a great idea! Please check this in.
— Reply to this email directly or view it on GitHub https://github.com/HazyResearch/bazaar/issues/7#issuecomment-127507203.
Will push my new wrapper functions (in the fabfile
) when done processing everything...
@raphaelhoffmann Especially since the loading time of coreNLP is so (relatively) long, it seems like a better default than
batch_size=1000
would be to divide lines by the number of cores (orcores*nodes
for distribute). For example I just got a 2x speedup on an ec-2 node really easily here (had forgotten to set batch size first time around...). Thoughts?