V2TrainingRequest

Body for POST /v2/training.

company_iduuidrequired

Company to train a policy for.

business_date object

Defaults to today.

anyOf

• date
• null

date

model_variantModelVariant

Which of the company's models to train, and so the horizon its states are assembled at. Each variant keeps its own weights, so training one leaves the other untouched. Defaults to the full-year model, which is what a company's single artifact was before the split.

Possible values: [H90, H365]

Default value: H365

epochsEpochs

Possible values: >= 1 and <= 200

Default value: 12

clip_epsilonClip Epsilon

Possible values: > 0 and <= 1

Default value: 0.2

learning_rateLearning Rate

Possible values: > 0 and <= 1

Default value: 0.05

auto_approveAuto Approve

When true, the freshly-trained model is also promoted to the approved slot in one call. When false, run POST /v2/model/approve (or set this flag on the next training call) to start using it for inference.

Default value: false

V2TrainingRequest

{
  "company_id": "3fa85f64-5717-4562-b3fc-2c963f66afa6",
  "business_date": "2024-07-29",
  "model_variant": "H365",
  "epochs": 12,
  "clip_epsilon": 0.2,
  "learning_rate": 0.05,
  "auto_approve": false
}