Metrics

These implement auxiliary model-learning objective functions or benchmarking evaluation criteria that are usable in hyper-tuning or model reporting goals

---

scikitlab.metrics.mutual_info.MutualInfoMetric

MutualInfoMetric(y=None, **kwargs)

Metric calculating the normalized mutual-information between random variables. This is a symmetric value indicating the 0-1 degree of dependence between two signals. This metric is more general than correlation in that can detect non-linear relationships without assumptions as well as work with continuous or discrete variables. This implementation wraps around scikit-learn's mutual info estimate calculations & can be treated as a learnable component to tune for hyperparameters.

Parameters:

Name	Type	Description	Default
y	Optional[array]	Vector of shape (n_samples,) to measure against. Useful to optimize computations when feature selecting against a constant target.	None
kwargs		other parameters as per scikits mutual_info_regression/classif	{}