wip modify_level(by) performance investigation

[nfultz]

Some benchmarking

# here we want at least one of each ideo st there aren't random failures in assn 9
draw_ideo <- function(N) {
  x <- c("Liberal", "Moderate", "Conservative")
  x <- c(x, sample(x, size=N-3, prob=c(.2,.3,.5), replace=TRUE))
  sample(x)
}

population <- function(N) fabricate(
  villages = add_level(
    N = N, elevation = rnorm(N),
    high_elevation = as.numeric(elevation > 0)
  ),
  individuals = add_level(N = 10, noise = rnorm(N)),
  individuals = modify_level(ideo_3 = draw_ideo(N), by = "villages")
)

population2 <- function(N) fabricate(
  villages = add_level(
    N = N, elevation = rnorm(N),
    high_elevation = as.numeric(elevation > 0)
  ),
  individuals = add_level(N = 10, noise = rnorm(N)),
  individuals = modify_level2(ideo_3 = draw_ideo(N), by = "villages")
)

require(profvis)
profvis(population())


microbenchmark::microbenchmark(population(100), population2(100),
                               population(200), population2(200),
                               population(300), population2(400),
                               population(800), population2(800),
                               population(1600), population2(1600))

> summary(lm(log(t)~log(N):which + which, df))

Call:
lm(formula = log(t) ~ log(N):which + which, data = df)

Residuals:
        Min          1Q      Median          3Q         Max 
-0.52720065 -0.15539903  0.00375551  0.17617271  1.23373942 

Coefficients:
                           Estimate  Std. Error    t value   Pr(>|t|)    
(Intercept)             -7.45114549  0.06281345 -118.62340 < 2.22e-16 ***
whichpopulation2         1.15412551  0.08955341   12.88757 < 2.22e-16 ***
log(N):whichpopulation   1.25744811  0.01044202  120.42193 < 2.22e-16 ***
log(N):whichpopulation2  1.02387495  0.01051372   97.38464 < 2.22e-16 ***
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

Residual standard error: 0.2304528 on 996 degrees of freedom
Multiple R-squared:  0.9603536,	Adjusted R-squared:  0.9602341 
F-statistic: 8042.017 on 3 and 996 DF,  p-value: < 2.2204e-16

[coveralls]

Coverage decreased (-3.3%) to 95.605% when pulling 8331507 on nfultz/modify_by_performance into fa31dab on master.