add support for etime.Expt

[etuleu]

This is to enable stats for multiple runs == one experiment.

I am actually not sure why we had the LastNRows shortcut at the end so I removed it. But please let me know if we need that back. I tested this in boa and seems to work fine but not sure what the downside might be if any.

[codecov]

Codecov Report

Patch coverage has no change and project coverage change: +0.01% 🎉

Comparison is base (2db038e) 35.13% compared to head (32d2fe6) 35.14%.

❗ Current head 32d2fe6 differs from pull request most recent head 0589e18. Consider uploading reports for the commit 0589e18 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##           master     #124      +/-   ##
==========================================
+ Coverage   35.13%   35.14%   +0.01%     
==========================================
  Files          80       80              
  Lines        6214     6211       -3     
==========================================
  Hits         2183     2183              
+ Misses       3824     3821       -3     
  Partials      207      207              

Files Changed	Coverage Δ
elog/stditems.go	0.00% <0.00%> (ø)

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[rcoreilly]

The LastNRows is for Run averaging over Epochs, and typically Run is the last one. I'll see about adding an option there.

[etuleu]

So if we don't do the LastN call it averages over all epochs right? Isn't that fine? Maybe we can make the LastN thing a special case if we need it.

[rcoreilly]

Yeah it is not ideal, but there are lots of calls to this method so I'm just going to push a version that always does the LastNRows thing for Run, and at least document that behavior. We could write an alternative version that doesn't do that, but in general you don't want to agg over all epochs because the start is learning, so in general this is better.