In addition to r, suppose we now have a validation r_validation (e.g. r_validation from another dataset or pseudo summary statistics). So we can train ghostbasil using r and compute validation R^2 based on r_validation as
Then can we stop basil when R2_validation stop increasing / reach its maximum? This is the current way that I choose optimal lambda and simulations show that it works well.
From an email discussion with Zihuai.