Author malte
Recipients andrew.coles, erez, florian, jendrik, malte, silvia
Date 2016-12-08.15:02:34
So far, so good! It would be nice to reduce the runtime penalty, but if it turns
out that we cannot, I don't think it would be a deal-breaker. Hopefully the
overhead will be much less noticeable with non-blind search settings.

I have two issues with the experiment:

1) If I understand correctly, the "base" numbers come from a separate
experiment. For the same reason that we randomize run IDs in our experiments, we
can get systematic errors if we combine data from two separate experiments. This
is less of an issue here than usually because these erros only affect runtime,
not memory usage. But I think we should still run a "clean" (all-in-one)
experiment before merging.

2) For a complete picture, we should have 32-bit and 64-bit data for all
configurations. That is, we should have data for v1-m32 and v2-m64, too.
Date User Action Args
2016-12-08 15:02:34maltesetmessageid: <>
2016-12-08 15:02:34maltesetrecipients: + malte, erez, andrew.coles, silvia, jendrik, florian
2016-12-08 15:02:34maltelinkissue213 messages
2016-12-08 15:02:34maltecreate