Add an intercept to the lm method

[rcannood]

Describe your changes

RcppArmadillo::fastLm(X, y) takes X as the design matrix verbatim -- unlike the formula interface, it does not add an intercept -- and predict.fastLm() then computes newdata %*% coef(object). Since lmds returns mean-centred coordinates, the model was effectively forced through the origin and could never fit the mean expression level of a gene.

Quick illustration, y = 5 + 2*x1 + noise on centred predictors:

> f  <- RcppArmadillo::fastLm(X, y)             # X has an intercept column
> head(predict(f, X), 3)
3.408 5.001 3.009
> f2 <- RcppArmadillo::fastLm(X[,-1], y)        # no intercept
> head(predict(f2, X[,-1]), 3)
-1.602 -0.009 -2.001    # truth: 3.606 4.982 3.017

This inflates the method's rmse and mae and depresses its per-cell correlations. Per-gene Pearson/Spearman are location-invariant, so those numbers were fine.

Adding the intercept column once, outside the per-gene loop, rather than in the pblapply() body.

Part of a series of PRs coming out of a pre-run review of the benchmark.

Checklist before requesting a review

• I have performed a self-review of my code

• Check the correct box. Does this PR contain:

  • Breaking changes
  • New functionality
  • Major changes
  • Minor changes
  • Bug fixes

• Proposed changes are described in the CHANGELOG.md

• CI Tests succeed and look good!