slow trends in spike counts

[mobeets]

residual CP not reliable because plotting Y vs Yh looks bad.

• [x] what's up with 20140304-MT-12 (just a bad cell accidentally fit)
• [x] fit history terms: regress Y(i) on Yh(i), Y(i-1), Y(i-2), ... 1
• [ ] also plot Y and Yh vs. binned dir strength, probably as boxScatterFitPlot

[mobeets]

looks like two lags is sufficient.

sometimes it does a difference (one weight positive, the other negative), sometimes they're both positive weights