非官方 serving;演示如何在自有服务中 embed leaves 批预测。
需要可拆独立仓的脚手架(优雅退出、max batch、/ready、热加载演示、Dockerfile)见
../serving-template(LIB-30)。本目录保持最小单文件 demo。
cd examples/http
export LEAVES_MODEL=../../testdata/xgboost_smoke.json
go run .
# 默认 :8080,可用 LEAVES_HTTP_ADDR 覆盖返回 ok。
单条(features 为一维特征向量):
{"features": [0,1,0,1,0,1,0,1]}响应:{"prediction": 0.42}
批预测(rows 或 batch,每行一条样本):
{
"rows": [
[0,1,0,1,0,1,0,1],
[1,0,1,0,1,0,1,0]
]
}响应:{"predictions": [0.42, 0.38], "nrows": 2}
扁平矩阵(features + nrows / ncols):
{
"features": [0,1,0,1,0,1,0,1, 1,0,1,0,1,0,1,0],
"nrows": 2,
"ncols": 8
}curl -s localhost:8080/predict -d '{"rows":[[0,1,0,1,0,1,0,1],[1,0,1,0,1,0,1,0]]}'