permit ISM binary ops on ISMs of different dimension?

[benthestatistician]

In some workflows one ISM ("aISM", say) might be built in the process of whittling down the sample toward a subsample that will be used for fitting an index score (eg propensity score) model, which will in turn be used to create another ISM ("bISM") for matching. (See statement of #142 for a bit more description.) In this context it would be helpful if when ismOpHandler saw ISMs e1 and e2 of different dimension, rather than quitting it just got rid of the rows and columns not common to both ISMs. Or something along those lines. As opposed to this:

> aglm = glm(formula = pr ~ . - cost, family = binomial, data = nuclearplants)
> mo1 <- caliper(aglm, width=1, data=nuclearplants)
> nuke.nopt <- subset(nuclearplants, pt==0)
> bglm <- update(aglm, data=nuke.nopt)
> mo2 <- match_on(bglm, caliper=1, within=mo1, data=nuke.nopt)
Error in ismOpHandler("+", e1, e2) : non-conformable matrices

(Speaking of #142, addressing this seems likely to require addressing it also.)

[benthestatistician]

Closing because:

• I no longer agree with my earlier suggestion to permit ISM binary ops when the ISMs don't share a dimension.
• I suspect (am not specifically confirming) that this has been addressed via #142, perhaps other changes. In particular,
• With the combination of subset methods for ISMs and Josh's new .[ methods for ISMs, one could address the problem in the example by subsetting mo1, down to observations found in nuke.nopt, before passing it to match_on() through the within= argument.