Contents
What are interaction features?
Feature interaction is a software engineering concept. It occurs when the integration of two features would modify the behavior of one or both features. The term feature is used to denote a unit of functionality of a software application.
What are interaction terms in logistic regression?
An interaction occurs if the relation between one predictor, X, and the outcome (response) variable, Y, depends on the value of another independent variable, Z (Fisher, 1926). Interactions are similarly specified in logistic regression if the response is binary.
What is automatic feature interaction?
AutoInt: Automatic Feature Interaction Learning via Self-Attentive Neural Networks. With different layers of the multi-head self-attentive neural networks, different orders of feature combinations of input features can be modeled. The whole model can be efficiently fit on large-scale raw data in an end-to-end fashion.
What does it mean when an interaction is not significant?
When there is no Significance interaction it means there is no moderation or that moderator does not play any interaction on the variables in question.
What does an interaction term mean?
In summary: When there is an interaction term, the effect of one variable that forms the interaction depends on the level of the other variable in the interaction. Without an interaction term, the mean value for Females on Med B would have been α+β1 +β2.
When to use h statistic in feature interaction?
That is what partial dependence plots are for. A meaningful workflow is to measure the interaction strengths and then create 2D-partial dependence plots for the interactions you are interested in. The H-statistic cannot be used meaningfully if the inputs are pixels.
How does the interaction statistic work in real life?
The interaction statistic works under the assumption that we can shuffle features independently. If the features correlate strongly, the assumption is violated and we integrate over feature combinations that are very unlikely in reality. That is the same problem that partial dependence plots have.
Which is a meaningful interpretation of the h-statistic?
The H-statistic has a meaningful interpretation: The interaction is defined as the share of variance that is explained by the interaction. Since the statistic is dimensionless, it is comparable across features and even across models. The statistic detects all kinds of interactions, regardless of their particular form.
How to generate interaction statistic under null hypothesis?
To generate the interaction statistic under the null hypothesis, you must be able to adjust the model so that it has no interaction between feature j and k or all others. This is not possible for all types of models. Therefore this test is model-specific, not model-agnostic, and as such not covered here.