2. Data cleaning#

The class ModifiedCausalForest has several data cleaning options to improve the estimation quality of your Modified Causal Forest. Below, you find a table with the relevant parameters and a brief description:

Parameter	Description
`dc_clean_data`	If True, all observations with missing values are dropped. Variables not required in the analysis are also removed. Default: True.
`dc_screen_covariates`	If True, covariates are screened and cleaned. Specifically, features without variation are dropped. Default: True.
`dc_check_perfectcorr`	If `dc_screen_covariates` is True, covariates perfectly correlated with others are removed. Default: True.
`dc_min_dummy_obs`	If `dc_screen_covariates` is True, binary (dummy) covariates with less than `dc_min_dummy_obs` zeroes or ones are removed. Default: 10.

Please consult the API for more details.

2.1. Example#

from mcf.example_data_functions import example_data
from mcf.mcf_functions 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", "x_cont1", "x_ord1"],
    # Parameters for data cleaning:
    dc_clean_data=True,
    dc_screen_covariates=True,
    dc_check_perfectcorr=False,
    dc_min_dummy_obs=100
)