Feat: Implement Kernel Riesz ATT Estimator with CV and DML

[apoorvalal]

This commit introduces a flexible KernelRieszATT estimator for Average Treatment Effect on the Treated (ATT) estimation using kernel-based Riesz representer methods.

Key features include:

• Support for various kernel functions (e.g., RBF, linear).
• Automatic hyperparameter selection for kernel parameters (e.g., gamma) and regularization strength via cross-validation, using Maximum Mean Discrepancy (MMD) as the default scoring metric.
• Optional Double Machine Learning (DML) through cross-fitting for more robust ATT estimates.
• Refactored existing ATT estimation methods into a common ATTWeightEstimator base class for better modularity.
• Updated simulation script to demonstrate and test the new estimator with its various configurations (CV, DML, different kernels).
• Added comprehensive docstrings and API refinements for all new and modified components.

[Copilot]

Pull Request Overview

This PR adds a new kernel-based Riesz ATT estimator with hyperparameter CV and optional DML, refactors existing ATT estimators into a shared base class, and provides an extended example notebook demonstrating all estimators.

• Introduce KernelRieszATT with kernel selection, CV-based tuning, and DML cross-fitting.
• Refactor penalized, constrained, and regression-form ATT estimators under ATTWeightEstimator.
• Provide an example notebook showing usage, parameter grids, and results comparison.

Reviewed Changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 2 comments.

File	Description
examples/riesz_representer.ipynb	Added a demo notebook showcasing KernelRieszATT and other estimators
cbpys/riesz_estimators.py	Implemented PenalizedFormEstimator, ConstrainedFormEstimator, and RieszRegressionFormEstimator
cbpys/kernel.py	Implemented KernelRieszATT with CV hyperparameter search and DML support
cbpys/baseclass.py	Added ATTWeightEstimator abstract base class

Comments suppressed due to low confidence (2)

cbpys/kernel.py:89

• [nitpick] The estimator currently only warns about unexpected kwargs like 'kernel_args' and 'reg_param'. Consider officially accepting these legacy aliases or raising a clear TypeError so users know the correct argument names.

            )

cbpys/kernel.py:9

• [nitpick] Add unit tests covering the CV parameter search (including edge cases), fallback to defaults, and DML cross-fitting logic to ensure all branches behave as intended.

class KernelRieszATT(ATTWeightEstimator):

[Copilot]

In the CV loop you fit weights on the validation splits instead of the training splits. To properly evaluate hyperparameters, use the training folds to fit (train_idx) and then score on the corresponding validation folds (val_idx).

Suggested change

	# Fit weights FOR THE VALIDATION FOLD using current params
	# These weights are specific to X_c_val_fold to balance X_t_val_fold
	weights_val_fold = self._fit_single_config(
	X_t_val_fold,
	X_c_val_fold,
	# Fit weights using training data
	weights_train_fold = self._fit_single_config(
	X_t_train_fold,
	X_c_train_fold,

[Copilot]

Use warnings.warn instead of print for warning messages so users can filter or redirect warnings appropriately.

Suggested change

	print(
	warnings.warn(