Interface: DmlAteRequest
@kortexya/reasoninglayer / DmlAteRequest
Interface: DmlAteRequest
Section titled “Interface: DmlAteRequest”Defined in: src/types/causal.ts:763
Request to estimate the ATE (or ATT) for continuous covariates via Double Machine Learning (cross-fitted AIPW).
Properties
Section titled “Properties”basisDegree?
Section titled “basisDegree?”
optionalbasisDegree?:number|null
Defined in: src/types/causal.ts:769
Polynomial basis degree for the nuisance models (default 1 = linear; 2 or
more captures non-linear confounding). Ignored when learner is
"forest" (trees use the raw features).
folds?
Section titled “folds?”
optionalfolds?:number|null
Defined in: src/types/causal.ts:771
Number of cross-fitting folds (default 5).
forestMaxDepth?
Section titled “forestMaxDepth?”
optionalforestMaxDepth?:number|null
Defined in: src/types/causal.ts:773
Maximum tree depth when learner is "forest" (default 8).
forestTrees?
Section titled “forestTrees?”
optionalforestTrees?:number|null
Defined in: src/types/causal.ts:775
Number of trees when learner is "forest" (default 100).
identification?
Section titled “identification?”
optionalidentification?:IdentificationRefDto|null
Defined in: src/types/causal.ts:780
Gate this estimation on a persisted identification certificate. Absent
means legacy behavior, honestly labeled caller_asserted.
learner?
Section titled “learner?”
optionallearner?:string|null
Defined in: src/types/causal.ts:786
Nuisance learner: "linear" (default — logistic + ridge over the
polynomial basis) or "forest" (a random forest on the raw features,
flexible under high-dimensional / non-smooth confounding).
observations
Section titled “observations”observations:
ContinuousObservationDto[]
Defined in: src/types/causal.ts:788
The observed sample with feature-vector covariates.
repetitions?
Section titled “repetitions?”
optionalrepetitions?:number|null
Defined in: src/types/causal.ts:795
Number of repeated cross-fitting partitions S (default 1 = a single
partition). With S >= 2 the estimate is the median over S seeded,
treatment-stratified partitions and its variance carries the
across-partition dispersion correction.
optionalseed?:number|null
Defined in: src/types/causal.ts:800
Seed for the repeated-cross-fitting partitions (default 0); results are
deterministic per (seed, repetitions).