Make xgboost optional in synthetic dataset generation

[Si-ra-kri]

This change makes xgboost an optional dependency for synthetic dataset generation. When xgboost is not installed, LinearRegression-based
learners continue to work without import errors.

[CLAassistant]

All committers have signed the CLA.

[Si-ra-kri]

I’ve fixed the indentation issue that caused the CI failure.
The latest commit resolves the error. Please let me know if you’d like me to trigger CI again.

[Si-ra-kri]

Hi! Just a gentle follow-up - all CI checks are passing now.
Please let me know if any changes are needed from my side.

[jeongyoonlee]

Code review

Found 1 issue:

1. Indentation bug in get_synthetic_preds silently drops S, T, and X learner predictions — only R Learner results are ever computed.

models = [(LinearRegression, "LR")] is correctly inside the for base_learner, label_l loop (line 76), but if XGBRegressor is not None: (line 77) and for model, label_m in models: (line 80) are dedented one level — placing them outside the outer loop. After all four base-learner iterations complete, base_learner and label_l hold only the last values (BaseRRegressor / "R"), so only "R Learner (LR)" (and "R Learner (XGB)" if installed) are added to preds_dict. The S, T, and X learner predictions are silently skipped.

causalml/causalml/dataset/synthetic.py

Lines 72 to 83 in ff8dea7

	for base_learner, label_l in zip(
	[BaseSRegressor, BaseTRegressor, BaseXRegressor, BaseRRegressor],
	["S", "T", "X", "R"],
	):
	models = [(LinearRegression, "LR")]
	if XGBRegressor is not None:
	models.append((XGBRegressor, "XGB"))
	for model, label_m in models:
	learner = base_learner(model())
	model_name = "{} Learner ({})".format(label_l, label_m)

get_synthetic_preds_holdout does not have this bug — all three lines sit correctly inside the outer loop there:

causalml/causalml/dataset/synthetic.py

Lines 380 to 390 in ff8dea7

	for base_learner, label_l in zip(
	[BaseSRegressor, BaseTRegressor, BaseXRegressor, BaseRRegressor],
	["S", "T", "X", "R"],
	):
	models = [(LinearRegression, "LR")]
	if XGBRegressor is not None:
	models.append((XGBRegressor, "XGB"))
	for model, label_m in models:
	# RLearner will need to fit on the p_hat
	if label_l != "R":
	learner = base_learner(model())

The fix is to indent lines 77–81 of get_synthetic_preds by four additional spaces so they match the structure in get_synthetic_preds_holdout.

🤖 Generated with Claude Code

_{- If this code review was useful, please react with 👍. Otherwise, react with 👎.}

[Si-ra-kri]

Thanks for catching that, indentation has been fixed and pushed.
Please let me know if anything else needs adjustment.

[jeongyoonlee]

LGTM