2.21 Booster attributes

Racket

2.21 Booster attributes🔗ℹ

A booster carries an arbitrary string→string attribute dictionary — handy for stamping a model with provenance. This example sets two attributes with booster-set-attr!, reads them back with booster-attr, lists the keys with booster-attr-names, then removes one with booster-delete-attr! (a missing key reads back as #f).

<r18-require> ::=

(require xgboost)

<r18-provide> ::=

(provide run-example)

<r18-run> ::=

(define (run-example)
  (define b (make-booster))
  (booster-set-attr! b "owner" "racket")
  (booster-set-attr! b "purpose" "example")
  (define before-delete
    (hash 'owner (booster-attr b "owner")
          'purpose (booster-attr b "purpose")
          'names (sort (booster-attr-names b) string<?)))
  (booster-delete-attr! b "purpose")
  (hash 'before-delete before-delete
        'purpose-after-delete (booster-attr b "purpose")
        'names-after-delete (booster-attr-names b)))

The harness "test/18-booster-attrs.rkt" prints the attributes and asserts the round-trip and the post-delete state.

<*> ::=

<r18-require>
<r18-provide>
<r18-run>

2.1	Building a DMatrix
2.2	Training a regressor
2.3	Binary classification
2.4	Multiclass classification
2.5	Watching an evaluation set
2.6	Iris: a full classification pipeline
2.7	Get Started
2.8	Robust regression
2.9	Quantile regression
2.10	Poisson count regression
2.11	Survival analysis (AFT)
2.12	Custom objective
2.13	Saving and loading models
2.14	Booster snapshots
2.15	DMatrix constructors
2.16	DMatrix metadata
2.17	Slicing and binary serialization
2.18	Quantile cuts
2.19	The high-level API end to end
2.20	Booster lifecycle and config
2.21	Booster attributes
2.22	Model dumps and feature importance
2.23	In-place prediction (dense)
2.24	In-place prediction (CSR)
2.25	In-place prediction (columnar)
2.26	Parameter recipes
2.27	Learning to rank
2.28	Global and process APIs
2.29	CUDA regression
2.30	CUDA classification