机器学习训练流程图

一张水平泳道式机器学习训练流程图，企业架构风格。三条水平泳道：顶部 「Experiment Log」（靛蓝）、中间 「Model」（青绿）、底部 「Infra」（石板灰）。Data 泳道：「Raw Data Ingestion」 → 「Data Validation」（菱形 「Schema OK?」 分支）→ 「Alert & Fix」 → "Train/Val/Test Split"。Model 泳道：「Feature Engineering」 → 「Train/Val/Test Split」（N epochs 循环箭头）→ 「Model Init」（菱形阈值判断）→ 「Training Loop」 → 「N epochs」。Infra 泳道：计算集群图标 → 「Validation」 → 「Metric ≥ threshold?」 → 「Save Checkpoint」。跨泳道虚线连接箭头。靛蓝、青绿、石板灰色系，白色背景，9pt 粗体标注，右下角图例。

A horizontal swimlane flowchart of a complete machine learning training pipeline, drawn in a clean enterprise-architecture style. Three horizontal swimlanes divide the chart: top lane labelled "Data" in indigo, middle lane labelled "Model" in teal, bottom lane labelled "Infra" in slate. Data lane: sequential process boxes — "Raw Data Ingestion" → "Data Validation" (with a diamond decision node checking "Schema OK?", with Yes/No branches: Yes continues, No routes to "Alert & Fix" box) → "Feature Engineering" → "Train/Val/Test Split" (three output arrows to labelled dataset icons). Model lane: "Model Init" → "Training Loop" (a loop-back arrow representing epochs, with a small gear icon and label "N epochs") → "Validation" (diamond "Metric ≥ threshold?", Yes → "Save Checkpoint", No → back to Training Loop) → "Hyperparameter Tuning" → "Final Evaluation". Infra lane: "Compute Cluster" icon → "Experiment Tracker" (labelled "MLflow"-style icon without using that brand name: just "Experiment Log") → "Model Registry" → "Deployment". Cross-lane connecting arrows in dashed lines. Colour palette: indigo, teal, slate on white. All box labels 9 pt bold. Legend at bottom-right.