7. Experimental features#

This section describes advanced and experimental features. Consult the method-specific API before combining them. Please open an issue here if you encounter any problems or have any questions.

7.1. Balancing Tests#

Treatment effects may be subject to selection bias if the distribution of the confounding features differs across treatment arms. The class ModifiedCausalForest provides the option to conduct balancing tests to assess whether the feature distributions are equal across treatment arms after adjustment by the Modified Causal Forest. The balancing tests are based on the estimation of average treatment effects (\(\text{ATE's}\)) with user-specified features as outcomes. If the features are balanced across treatment arms, the estimated \(\text{ATE's}\) should be close to zero.

The Modified Causal Forest runs balancing tests for the features specified in the parameters var_x_name_balance_test_ord and var_x_name_balance_test_unord if the parameter p_bt_yes is set to True. See also the table below.

Parameter	Purpose
`p_bt_yes`	Enable balancing tests.
`var_x_name_balance_test_ord`	Ordered variables used as balancing-test outcomes.
`var_x_name_balance_test_unord`	Unordered variables used as balancing-test outcomes.

Please consult the API for more details.

The results of the balancing tests are part of the txt-file in the output folder that the mcf package generates. You can find the location of this folder by accessing the “outpath” entry of the gen_cfg attribute of your Modified Causal Forest:

from mcf.example_data import example_data
from mcf.mcf_main import ModifiedCausalForest

# Generate example data using the built-in function `example_data()`
training_df, prediction_df, name_dict = example_data()

my_mcf = ModifiedCausalForest(
    var_y_name="outcome",
    var_d_name="treat",
    var_x_name_ord="x_cont0"
    )

my_mcf.gen_cfg.outpath

7.1.1. Example#

from mcf.example_data import example_data
from mcf.mcf_main import ModifiedCausalForest

# Generate example data using the built-in function `example_data()`
training_df, prediction_df, name_dict = example_data()

params = dict(
    var_y_name="outcome",
    var_d_name="treat",
    var_x_name_ord=["x_cont0", "x_cont1", "x_ord1"],
    var_x_name_unord=["x_unord0"],
    # Parameters for balancing tests:
    p_bt_yes=True,
    var_x_name_balance_test_ord=["x_cont0", "x_cont1", "x_ord1"],
    var_x_name_balance_test_unord=["x_unord0"]
)

my_mcf = ModifiedCausalForest(**params)
my_mcf.train(training_df)
results = my_mcf.predict(prediction_df)

# sensitivity() needs a new, untrained instance with the same settings.
my_mcf_sens = ModifiedCausalForest(**params)
sensitivity_results = my_mcf_sens.sensitivity(
    training_df, prediction_df, results=results
)

7.2. Sensitivity checks#

sensitivity() provides a simulation-based diagnostic. The implemented basic scenario estimates treatment probabilities, selects a reference treatment population and simulates placebo treatments. Effects are known to be zero in this scenario, so estimated deviations from zero help assess the procedure in that simulated setting. Replications reduce simulation noise. This does not identify or bound arbitrary unobserved confounding.

The following fragment uses an existing mcf object and data:

sensitivity_results = my_mcf.sensitivity(
    training_df, prediction_df, results=results
)

results is required; pass None if no prior results are to be used. When supplied results contain an IATE dataframe and name metadata, those rows take precedence over prediction_df. Feature selection is not supported by this sensitivity procedure. See the method API for compatible settings.

7.3. GPU Support#

Set _int_cuda=True to request GPU processing. This requires CUDA-capable PyTorch and may fall back to the CPU when CUDA is unavailable. The IATE GPU path also requires low-memory prediction and bias adjustment to be disabled, and fewer than 16 selected IATE workers. Availability does not guarantee a speed improvement; see the API for current restrictions.