Debug causalBoosting in the observational case

[saberpowers]

As pointed out by William Nilsson in email correspondence, when a stratum contains only treatment observations or only control observations, causalBoosting() encounters a fatal error. This is presumably true of causalMARS() and PTOforest() as well. The issue here is to check at the beginnings of these functions that each stratum contains both treatment and control observations, and to exit with an informative error message if not.

[saberpowers]

Indeed the same fatal error can occur even if all strata have both treatment and control observations.

For example:

> n = 100 # number of training-set patients to simulate
> p = 10  # number of features for each training-set patient
> 
> # Simulate data
> x = matrix(rnorm(n * p), nrow = n, ncol = p) # simulate covariate matrix
> tx_effect = x[, 1] + (x[, 2] > 0) # simple heterogeneous treatment effect
> tx = rbinom(n, size = 1, p = 0.5) # random treatment assignment
> y = rowMeans(x) + tx * tx_effect + rnorm(n, sd = 0.001) # simulate response
> set.seed(1)
> stratum = sample(rep(1:2, length = n))
> table(tx, stratum)
#   stratum
#tx   1  2
#  0 23 21
#  1 27 29
> fit_cb = causalLearning::causalBoosting(x, tx, y, num.trees = 500, propensity = TRUE, stratum = stratum)
#R(8164,0x7fff7b266300) malloc: *** error for object 0x7fa603d23a00: pointer being freed was not #allocated
#*** set a breakpoint in malloc_error_break to debug
#Abort trap: 6

So I'm going to re-name this issue: Debug causalBoosting in the observational case. Given that we ran causalBoosting in the observational case many times in the course of writing the paper, I suspect that this bug was introduced by one of these commits:

• 7601b55
• 8f00fae

[Christinawang666]

Hi, when I run causalBoosting with propensity=TRUE, the same fatal error happens and I wonder if the problem has been solved already?

[saberpowers]

I'm sorry to report that it has not been solved.