AGENT SKILL

mlops-observability-cn

Full stack observability - reproducibility, lineage, monitoring, alerting

License 许可
MIT
最近更新
2026-06-10
Stars
★ 3
Forks
0

01 前言 Front Matter

name
mlops-observability-cn
license
MIT
version
1.0.0
description
Full stack observability - reproducibility, lineage, monitoring, alerting

03 SKILL.md 正文

MLOps Observability 👁️

Glass box system - reproducible, traceable, monitored.

Features

1. MLflow Tracking 📊

Complete tracking setup:

cp references/mlflow-tracking.py ../your-project/src/tracking.py

Tracks:

  • Config (params)
  • Metrics (accuracy, loss)
  • Models (sklearn/pytorch)
  • Datasets (lineage)
  • Git commit (reproducibility)

2. Drift Detection 📉

Using Evidently:

from evidently import Report
from evidently.metrics import DataDriftTable

report = Report(metrics=[DataDriftTable()])
report.run(reference_data=train, current_data=prod)

3. Explainability (SHAP) 🔍

import shap

explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X)
shap.summary_plot(shap_values, X)

Quick Start

# Copy tracking code
cp references/mlflow-tracking.py ./src/

# Add to training script:
# from tracking import setup_tracking, log_training_run

Reproducibility

# Set all seeds
import random, numpy as np, torch
random.seed(42)
np.random.seed(42)
torch.manual_seed(42)

# Track git commit
import git
commit = git.Repo().head.commit.hexsha
mlflow.log_param("git_commit", commit)

Monitoring Checklist

  • [ ] Random seeds fixed
  • [ ] MLflow tracking enabled
  • [ ] System metrics logged
  • [ ] Drift detection setup
  • [ ] SHAP explanations saved
  • [ ] Alerts configured

Alerting

  • Local: plyer notifications
  • Production: PagerDuty (critical) / Slack (warnings)

Author

Converted from MLOps Coding Course

Changelog

v1.0.0 (2026-02-18)

  • Initial OpenClaw conversion
  • Added MLflow tracking code

04 同仓库更多 Skills