Fix off-by-1 indexing bug (#154)

[mooreryan]

Take EMburn = 3, and EMiter = 6. Originally, we're indexing from (3+1):(6+1) -> 4:7 -> c(4, 5, 6, 7). The first item is the init value. The next three c(2, 3, 4) are in the burn section, and the final three c(5, 6, 7) are in the keep section. However, we're pulling in the final one in the burn section. So it should be EMburn + 2.

See #154.

---

I assume the actual numerical difference made by this change would be small. To do a quick check of that I ran through the debugger using this data:

data(Lee)
Lee_phylum <- phyloseq::tax_glom(Lee, taxrank="Phylum")
divnet_phylum_char <- divnet(Lee_phylum, X = "char", tuning = "test")

Then added some diagnostic plots to the fit_atichison function. Something like this:

b0_EM:

  b0_EM <- colMeans(b0_list[(EMburn + 2):(EMiter + 1), ])

  b0_EM_old <- colMeans(b0_list[(EMburn + 1):(EMiter + 1), ])
  plot(b0_EM, b0_EM_old)
  abline(a = 0, b = 1)

b_EM:

    b_list_reduced <- b_list[,, (EMburn + 2):(EMiter + 1)]

    b_list_reduced_old <- b_list[,, (EMburn + 1):(EMiter + 1)]

    if (length(dim(b_list_reduced)) == 2) {
      b_list_reduced <- b_list_reduced %>% array(c(1, dim(b_list_reduced)))
      b_list_reduced_old <- b_list_reduced_old %>%
        array(c(1, dim(b_list_reduced_old)))
    }

    b_EM <- apply(b_list_reduced, c(1, 2), mean)
    b_EM_old <- apply(b_list_reduced_old, c(1, 2), mean)
    output_list$beta <- b_EM

    plot(as.vector(b_EM), as.vector(b_EM_old))
    abline(a = 0, b = 1)

sigma_em:

  sigma_em <- apply(sigma_list[,, (EMburn + 2):(EMiter + 1)], c(1, 2), mean)

  sigma_em_old <- apply(sigma_list[,, (EMburn + 1):(EMiter + 1)], c(1, 2), mean)
  plot(as.vector(sigma_em), as.vector(sigma_em_old))
  abline(a = 0, b = 1)

(If you look at the actual values there very close.)

[adw96]

LGTM, thanks so much for this fix, @mooreryan