A sensible metalearner is chosen based on the outcome type.
default_metalearner(outcome_type)a Variable_Type object
For binary and continuous outcome types, the default metalearner is
non-negative least squares (NNLS) regression (Lrnr_nnls), and
for others the metalearner is Lrnr_solnp with an appropriate
loss and combination function, shown in the table below.
| Outcome Type | Combination Function | Loss Function |
| categorical | metalearner_linear_multinomial | loss_loglik_multinomial |
| multivariate | metalearner_linear_multivariate | loss_squared_error_multivariate |