MAS
A model-agnostic variable selection method that requires no assumption on the covariate distribution. It achieves FDR control through agent-X importance and cross-fitting.
Timeline (last 90 days)
A new study proposes a complementary method called MAS (Model-Agnostic Selection with agent-X importance) for achieving FDR-controlled variable selection without assumptions on the covariate distribution.
The core of MAS is the definition of an 'agent-X importance' measure, which evaluates at the population level the difference in 'oracle predictability' caused by the presence or absence of a certain variable.
MAS introduces a 'resplit-per-covariate' strategy, which decouples the dependencies among estimators by independently resampling for each covariate.
MAS enhances detection power through data aggregation techniques without sacrificing FDR control.
Under mild conditions, MAS establishes a theoretical guarantee of asymptotic FDR control.
Simulation experiments and real data analysis show that MAS performs comparably to or even better than existing methods in all three dimensions: FDR control, computational efficiency, and statistical power.
MAS and Model-X knockoffs are complementary: when the covariate distribution is known or can be estimated accurately, knockoffs may be the more efficient choice; when the distribution is unknown or difficult to model, MAS exhibits a model-agnostic advantage.
All Facts (7)
MAS enhances detection power through data aggregation techniques without sacrificing FDR control.
50%UnverifiedThe core of MAS is the definition of an 'agent-X importance' measure, which evaluates at the population level the difference in 'oracle predictability' caused by the presence or absence of a certain variable.
50%UnverifiedMAS introduces a 'resplit-per-covariate' strategy, which decouples the dependencies among estimators by independently resampling for each covariate.
50%UnverifiedMAS and Model-X knockoffs are complementary: when the covariate distribution is known or can be estimated accurately, knockoffs may be the more efficient choice; when the distribution is unknown or difficult to model, MAS exhibits a model-agnostic advantage.
50%UnverifiedUnder mild conditions, MAS establishes a theoretical guarantee of asymptotic FDR control.
50%UnverifiedSimulation experiments and real data analysis show that MAS performs comparably to or even better than existing methods in all three dimensions: FDR control, computational efficiency, and statistical power.
50%UnverifiedA new study proposes a complementary method called MAS (Model-Agnostic Selection with agent-X importance) for achieving FDR-controlled variable selection without assumptions on the covariate distribution.
50%