CausalEffect task performs causal inference analysis to estimate the causal effects of treatments on target variables using machine learning methods.

This task implements causal effect estimation using LinearDML for discrete treatments and CausalForestDML for continuous treatments. It analyzes all possible combinations of the specified target variables and estimates the Average Treatment Effect (ATE) for each treatment-target pair.

What this task does:

• Identifies treatment from your dataset
• For each treatment-target combination, estimates the causal effect using appropriate DML models
• Handles both discrete and continuous treatment variables
• Uses feature selection based on mutual information to select relevant confounders
• Generates results for all possible combinations of target variables
• Creates heatmap visualizations of causal effects
• Exports results as CSV files and PNG heatmaps

Input Requirements:

• Data: A Table containing numerical data with:
  • At least one target variable
  • At least one treatment variable
  • Optional: Additional variables that serve as confounders/covariates
  • All variables should be numerical
  • Missing values will be handled automatically

Configuration:

• targets: List of column names from your data that represent target variables These are the variables for which you want to measure causal effects

Outputs:

• results_folder: A folder containing:
  • A subfolder for each combination of target variables
    • Each subfolder contains:
    • CSV file with causal effect estimates for each treatment-target pair
    • Heatmap visualization of the causal effects for that combination

Example Use Case: If you have data with variables like 'drug_dose', 'exercise_hours', 'blood_pressure', 'cholesterol' and you specify targets=['blood_pressure', 'cholesterol'], the task will:

1. Treat 'drug_dose' and 'exercise_hours' as treatments
2. Estimate how each treatment affects each target
3. Generate results for individual targets and their combination
4. Create visualizations showing the strength and direction of causal effects

Note: The task uses sophisticated causal inference methods that account for confounding variables automatically, but results should be interpreted carefully considering domain knowledge and potential unmeasured confounders.