Reductions¶
The concept of a reduction is to take a complex problem and reduce it to a simpler problem. In ReductionML these are atomic units of functionality that are typed based on the features, label and prediction types they consume and produce. Consume means the type that is passed to it and produce is the type is passes to the next reduction in the chain.
For example, contextual bandit exploration algorithms reduce to contextual bandit scorers which in turn use regression.
Kinds¶
Linear regressors¶
Parameter free linear regressor
base reduction
Contextual bandit exploration¶
Explore based on the \(epsilon\) value
cb exploration
SquareCB exploration algorithm
cb exploration
Explore based on a softmax distribution of the predicted action scores
cb exploration
Contextual bandit scorer¶
Contextual bandit scorer for action dependent features
Other¶
Generate features as the elementwise multiplication of two vectors