2.23 In-place prediction (dense)
For serving, building a DMatrix per request is wasteful. predict-from-dense runs prediction directly on a row-major f32vector, skipping the DMatrix entirely. This example trains a regressor and checks the in-place result matches the DMatrix path exactly.
<r20-require> ::=
(require ffi/vector xgboost)
<r20-provide> ::=
(provide run-example)
Helpers. A trained booster and an approximate f32vector equality:
<r20-helpers> ::=
(define features (f32vector 1.0 2.0 0.5 2.0 1.0 1.5 3.0 0.5 0.0 0.5 3.0 2.0 4.0 2.0 1.0 1.5 1.5 0.5 2.5 3.5 1.5 0.0 1.0 0.0)) (define labels (f32vector 3.5 3.5 6.5 2.0 9.0 4.0 7.0 1.0)) (define (f32vector~= a b) (and (= (f32vector-length a) (f32vector-length b)) (for/and ([i (in-range (f32vector-length a))]) (< (abs (- (f32vector-ref a i) (f32vector-ref b i))) 1e-6)))) (define (make-trained) (define dtrain (make-dmatrix features #:nrow 8 #:ncol 3 #:missing -1.0 #:labels labels)) (values (train dtrain #:objective "reg:squarederror" #:max-depth 3 #:eta 0.1 #:verbosity 0 #:rounds 20) dtrain))
<r20-run> ::=
(define (run-example) (define-values (booster dtrain) (make-trained)) (define dmatrix-preds (predict booster dtrain #:as 'f32vector)) (define inplace-preds (predict-from-dense booster features #:nrow 8 #:ncol 3 #:missing -1.0 #:as 'f32vector)) (hash 'prediction-count (f32vector-length inplace-preds) 'inplace-preds inplace-preds 'dmatrix-preds dmatrix-preds 'matches-dmatrix? (f32vector~= inplace-preds dmatrix-preds)))
The harness "test/20-inplace-predict-dense.rkt" prints the prediction count and asserts the in-place predictions equal the DMatrix predictions.
<*> ::=