Estimation Robustness

The benefit of this estimator over the standard MLE estimator is that it comes with robustness guarantees to outliers or contaminated data. As an example, we estimate the mean for a Gaussian random variable.

>>> from regmmd import MMDEstimator
>>> from regmmd.utils import print_summary
>>>
>>> x = rng.normal(loc=0, scale=1.5, size=(50,))
>>>
>>> # We contaminate  only one sample
>>> x[42] = 100
>>>
>>> mmd_estim = MMDEstimator(
...     model="gaussian",
...     par_v=None,
...     par_c=None,
...     kernel="Gaussian",
...     solver={"type": "GD",
...             "burnin": 500,
...             "n_step": 1000,
...             "stepsize": 1,
...             "epsilon": 1e-4
...      },
...      random_state=42
... )
>>> res = mmd_estim.fit(X=x)
>>> print(np.mean(x))
2.153118443631034
>>> print(res["estimator"][0])
0.07003082198862483