Design and execute A/B tests for ML models in production using traffic splitting, statistical significance testing, and canary/shadow deployment strategies. Measure performance differences and make data-driven decisions about model rollout. Use when validating a new model version before full rollout, comparing candidate models trained with different algorithms, measuring business metric impact of model changes, or when regulatory requirements mandate gradual rollout.
Scanned 9/3/2026
Install to Claude Code
npx -y skills add pjt222/agent-almanac --skill run-ab-test-models --agent claude-codeInstalls into .claude/skills of the current project.
Are you the author of Run Ab Test Models?
Add the live security badge to your README — it updates automatically with every re-scan.
[](https://www.skillsdirectory.com/skills/pjt222-run-ab-test-models-ac416959)More formats (shields.io, HTML) on the badges page.
---
name: run-ab-test-models
locale: wenyan-ultra
source_locale: en
source_commit: 82c77053
fence_basis_commit: 82c77053
translator: "Julius Brussee homage — caveman"
translation_date: "2026-05-03"
description: >
Design and execute A/B tests for ML models in production using traffic splitting,
statistical significance testing, and canary/shadow deployment strategies. Measure
performance differences and make data-driven decisions about model rollout. Use when
validating a new model version before full rollout, comparing candidate models trained
with different algorithms, measuring business metric impact of model changes, or when
regulatory requirements mandate gradual rollout.
license: MIT
allowed-tools: Read Write Edit Bash Grep Glob
metadata:
author: Philipp Thoss
version: "1.0"
domain: mlops
complexity: intermediate
language: multi
tags: ab-testing, canary, shadow-deployment, traffic-splitting, statistical-significance, experimentation
---
# 行 A/B 模測
> 全配與板見 [Extended Examples](references/EXAMPLES.md)。
於產分流、計析、金/影釋以較模本。
## 用
- 釋新模本欲驗於全→用
- 較異算/特之候模→用
- 試超參改於業指影→用
- 量產上模效不冒全流→用
- 監管漸釋(如醫 ML)→用
- 評費效衡(模大)→用
## 入
- **必**:冠模(當產本)
- **必**:挑模(試新本)
- **必**:流分比(如 5% 予挑)
- **必**:成指(業與 ML)
- **必**:最小樣或測時
- **可**:護指(延、誤率限)
- **可**:用段(分層測)
## 行
### 一:設驗
定測參、成準、計需。
```python
# ab_test/experiment_config.py
from dataclasses import dataclass
from typing import List, Dict
import numpy as np
from scipy.stats import norm
@dataclass
# ... (see EXAMPLES.md for complete implementation)
```
得:驗配含計健全之樣大算,常 5-10k/變於 5-10% MDE。
敗:樣需過大→增分、延時、納大 MDE;驗基指準;考序測續察。
### 二:行流分
設路邏隨配請於模。
```python
# ab_test/traffic_router.py
import hashlib
import random
from typing import Dict, Optional
from dataclasses import dataclass
import logging
logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)
```
得:用-變一致配、流分準合配比、諸配記以析。
敗:驗散函生均勻(試 10k user_id)、查 user_id 跨請求穩(非 session_id)、確日誌捕諸測事件、首 1000 請驗分。
### 三:行影釋(可)
並行挑模而不擾用(影模)。
```python
# ab_test/shadow_deployment.py
import asyncio
from typing import Dict, Any
import logging
from concurrent.futures import ThreadPoolExecutor
import time
logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)
```
得:冠測常延供、挑測異步記不阻、測異捕以析。
敗:挑超時 < 冠 SLA、優雅理挑誤、察記憶(兩模載)、考採樣(記 10% 影測)。
### 四:採析指
集驗資、行計測。
```python
# ab_test/analysis.py
import pandas as pd
import numpy as np
from scipy import stats
from typing import Dict, Tuple
import logging
logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)
```
得:計測果含 p、信區、明決(釋/留/未定),常 7-14 日後或達樣。
敗:驗真標可(或需延析)、查樣比錯(SRM)示配漏、足樣達、察初新/首因、考序測若定平太緩。
### 五:察護指
續查挑不破安限。
```python
# ab_test/guardrails.py
import pandas as pd
import logging
from typing import Dict, List
logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)
```
得:護違 5-15 分內察、自停若關限破(延、誤)、警送組。
敗:驗護限現實(不過嚴)、確察循環續行、查 stop_experiment() 真更路、測警送。
### 六:作釋決
按驗果決挑釋否。
```python
# ab_test/rollout_decision.py
import logging
from typing import Dict
from dataclasses import dataclass
logger = logging.getLogger(__name__)
# ... (see EXAMPLES.md for complete implementation)
```
得:明決(全/漸釋、留冠、延測)含理與行項。
敗:決不明→行子組析(按段、時、機)、查互效、覆業境(2% 升值工本乎?)、徵相關方。
## 驗
- [ ] 流分合配比(內 1%)
- [ ] 同用恆配同變(一致查)
- [ ] 樣大算合理(5-50k/變)
- [ ] 計測 p 合手算
- [ ] 護違 5 分內警
- [ ] 影釋示模測異 < 5%
- [ ] 驗報含信區
- [ ] 釋決文錄附理
## 忌
- **樣比錯(SRM)**:察分異於配(95/5 變 92/8)→配漏;查散函均
- **窺**:未達樣前查果脹一型誤;用序測或待定終
- **新效**:用初應殊;行 ≥ 2 週見穩態
- **承效**:前變露擾今;用新用或足洗期
- **多測**:多指增假陽;以 Bonferroni 正或重一主指
- **力不足**:小流配需月察實效;平衡計力與險忍
- **忽段**:聚升可藏要段負影;行子組析
- **歸誤**:確指正歸於模測(非他系變)
## 參
- `deploy-ml-model-serving`
- `monitor-model-drift`
Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.
No comments yet. Be the first to comment!