---

name: ds-ab-testing compatibility: opencode completeness: 95 content-types:

• code
• guidance
• do-dont
• examples description: Provides Designs and analyzes A/B tests including hypothesis testing power analysis, sample size calculation, and statistical significance evaluation license: MIT maturity: stable metadata: domain: coding output-format: code related-skills: ds-classification-metrics, ds-experimental-design, ds-hypothesis-testing ds-metrics-and-kpis ds-statistical-power role: implementation scope: implementation triggers: A/B testing, A/B test, statistical test, power analysis, sample size how do I design tests, unit tests, testing archetypes:
  • tactical
  • generation anti_triggers:
  • brainstorming
  • vague ideation
  • code golf
  • over-engineering response_profile: verbosity: low directive_strength: high abstraction_level: operational version: "1.0.0"

---

A/B Testing

Comprehensive guide to a/b testing in machine learning and data science workflows.

When to Use This Skill

• Solving real-world experimentation & a/b testing problems
• Building machine learning pipelines with a/b testing
• Implementing best practices for a/b testing
• Optimizing model performance using a/b testing techniques
• Learning industry-standard approaches to a/b testing

When NOT to Use This Skill

• When using pre-built libraries without understanding underlying concepts
• For toy problems that don't require a/b testing rigor
• When domain expertise in specific problem requires different approach
• If your problem doesn't require the complexity this skill provides

Purpose and Key Concepts

A/B Testing is a critical component of the machine learning workflow. This skill covers:

1. Theoretical foundations — Mathematical principles and statistical concepts
2. Practical implementation — Working code examples and patterns
3. Common pitfalls — Mistakes to avoid and how to recover from them
4. Best practices — Industry-standard approaches and optimization techniques

Core Workflow

1. Understand the problem — Clearly define what you're solving for
2. Select approach — Choose the right technique for your data and constraints
3. Implement solution — Write clean, tested code following best practices
4. Validate results — Verify your implementation with tests and validation
5. Optimize performance — Improve efficiency and accuracy incrementally

Implementation Patterns

Pattern 1: Basic A/B Testing

import pandas as pd
import numpy as np
from scipy import stats

def run_basic_ab_test(group_a: pd.Series, group_b: pd.Series, alpha: float = 0.05) -> dict:
    """
    Perform a two-sample t-test to compare means between two groups.
    Returns p-value, confidence interval, and effect size.
    """
    if len(group_a) < 2 or len(group_b) < 2:
        raise ValueError("Each group must contain at least 2 observations.")

    t_stat, p_value = stats.ttest_ind(group_a, group_b, equal_var=False)
    mean_diff = group_b.mean() - group_a.mean()
    pooled_std = np.sqrt(((len(group_a) - 1) * group_a.var() + (len(group_b) - 1) * group_b.var()) / (len(group_a) + len(group_b) - 2))
    cohens_d = mean_diff / pooled_std if pooled_std > 0 else 0.0

    se_diff = np.sqrt(group_a.var()/len(group_a) + group_b.var()/len(group_b))
    ci_lower = mean_diff - stats.t.ppf(1 - alpha / 2, df=len(group_a) + len(group_b) - 2) * se_diff
    ci_upper = mean_diff + stats.t.ppf(1 - alpha / 2, df=len(group_a) + len(group_b) - 2) * se_diff

    return {
        "p_value": float(p_value)
        "significant": bool(p_value < alpha)
        "mean_difference": float(mean_diff)
        "confidence_interval": (float(ci_lower), float(ci_upper))
        "cohens_d": float(cohens_d)
        "group_a_mean": float(group_a.mean())
        "group_b_mean": float(group_b.mean())
    }

Pattern 2: Production-Ready A/B Testing

import logging
import pandas as pd
import numpy as np
from scipy import stats
from statsmodels.stats.power import TTestIndPower
from typing import Any, Dict, Optional

logger = logging.getLogger(__name__)

class ABTestAnalyzer:
    """Production-grade A/B test analyzer with power analysis and validation."""
    
    def __init__(self, alpha: float = 0.05, min_effect_size: float = 0.2):
        self.alpha = alpha
        self.min_effect_size = min_effect_size
        
    def validate_data(self, df: pd.DataFrame, group_col: str, metric_col: str) -> None:
        if df.empty:
            raise ValueError("Input DataFrame is empty.")
        if group_col not in df.columns or metric_col not in df.columns:
            raise ValueError(f"Missing required columns: {group_col}, {metric_col}")
        if df[metric_col].dtype not in [np.float64, np.int64]:
            raise ValueError("Metric column must be numeric.")
            
    def analyze(self, df: pd.DataFrame, group_col: str, metric_col: str) -> Dict[str, Any]:
        self.validate_data(df, group_col, metric_col)
        group_a = df.loc[df[group_col] == 0, metric_col].dropna()
        group_b = df.loc[df[group_col] == 1, metric_col].dropna()
        
        if len(group_a) < 30 or len(group_b) < 30:
            logger.warning("Sample sizes below 30. Results may be unreliable.")
            
        t_stat, p_value = stats.ttest_ind(group_a, group_b, equal_var=False)
        mean_diff = group_b.mean() - group_a.mean()
        pooled_std = np.sqrt(((len(group_a) - 1) * group_a.var() + (len(group_b) - 1) * group_b.var()) / (len(group_a) + len(group_b) - 2))
        cohens_d = mean_diff / pooled_std if pooled_std > 0 else 0.0
        
        power_analysis = TTestIndPower()
        required_n = power_analysis.solve_power(effect_size=abs(cohens_d), alpha=self.alpha, power=0.8, ratio=1.0)
        
        return {
            "p_value": float(p_value)
            "significant": bool(p_value < self.alpha)
            "mean_difference": float(mean_diff)
            "effect_size_cohens_d": float(cohens_d)
            "required_sample_size_per_group": int(np.ceil(required_n))
            "actual_sample_sizes": {"group_a": len(group_a), "group_b": len(group_b)}
            "status": "pass" if p_value < self.alpha and abs(cohens_d) >= self.min_effect_size else "inconclusive"
        }

Best Practices

• ✅ Always validate your implementation on test data
• ✅ Document your assumptions and methodology
• ✅ Use version control for reproducibility
• ✅ Monitor performance metrics in production
• ✅ Periodically review and update your approach
• ✅ Test with edge cases and outliers
• ✅ Log all significant operations for debugging

Common Pitfalls

| Pitfall | Problem | Solution | |

---

---

Constraints

MUST DO

• Validate all data preprocessing steps are fit-only on training data, never on validation or test sets
• Implement reproducible pipelines with fixed random seeds and deterministic operations where possible
• Report model performance with confidence intervals via bootstrapping or cross-validation across multiple runs
• Log all experiments with parameters, metrics, and artifacts using MLflow or equivalent tracking system

MUST NOT DO

• Do not evaluate a model on the same data used for training — always hold out a proper test set
• Avoid overfitting to the validation set by limiting hyperparameter search iterations
• Never use features that can only be computed at inference time (look-ahead bias)
• Do not report single-run accuracy without statistical significance testing or error bars

Live References

Authoritative documentation links for this skill's domain. The model follows markdown links at load time to resolve external references and inline content.

• A/B Testing — Wikipedia
• Google Analytics Statistics Guide
• Scikit-learn Model Evaluation
• Statistical Significance Calculator (Real Statistics)
• Optimizely A/B Testing Best Practices