# Template for data science blog posts
blog_template = """
# [Action-oriented title]
## e.g., "How I Reduced Model Latency by 10x Without Sacrificing Accuracy"

## Introduction (100-150 words)
- Hook: Start with a surprising result or relatable problem
- Context: What's the real-world scenario?
- Promise: What will the reader learn?

## The Problem (200-300 words)
- Business context
- Why existing solutions fall short
- Constraints and requirements

## The Approach (500-800 words)
- Solution overview with architecture diagram
- Key design decisions and tradeoffs
- Code snippets (keep them short and focused)

## Results (300-500 words)
- Quantitative metrics with before/after
- Visualizations
- Ablation studies if applicable

## Lessons Learned (200-300 words)
- What worked
- What didn't
- What you'd do differently

## Next Steps (100 words)
- Future improvements
- Open questions
- Links to resources
"""

# Cell structure for effective notebooks

# Cell 1: Title and context
"""
# Predicting Customer Churn with Gradient Boosting

**Business Goal:** Identify at-risk customers for targeted retention campaigns
**Dataset:** 50K customer records with 6 months of behavioral data
**Result:** 89% AUC, $2.3M estimated annual impact
"""

# Cell 2: Setup and imports
"""
## Setup
"""
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import seaborn as sns

# Cell 3: Data exploration with narrative
"""
## What does our data look like?
"""
# Always explain what you're looking for
df = pd.read_csv('customers.csv')

# Add markdown before visualization
"""
The distribution of customer tenure shows a bimodal pattern,
suggesting two distinct customer segments. This insight will
inform our feature engineering.
"""
plt.figure(figsize=(10, 6))
plt.hist(df['tenure_months'], bins=30, edgecolor='black')
plt.xlabel('Tenure (months)')
plt.ylabel('Count')
plt.title('Customer Tenure Distribution')
plt.show()

# Cell 4: Feature engineering with reasoning
"""
## Feature Engineering

Key insight: Recent behavior is more predictive than historical averages.
We create rolling window features at 7, 30, and 90-day horizons.
"""

# Cell 5: Model training with clear output
"""
## Model Training

We use LightGBM for its speed and native handling of categorical features.
"""
import lightgbm as lgb

model = lgb.LGBMClassifier(n_estimators=100, max_depth=6)
model.fit(X_train, y_train)

# Cell 6: Evaluation with business context
"""
## Results

Our model achieves 89% AUC, meaning it correctly ranks a churner higher 
than a non-churner 89% of the time. In practical terms:

- **Top 10% flagged customers** have a 45% actual churn rate (vs 8% baseline)
- **Expected impact:** Retaining 30% of flagged customers = $2.3M/year
"""

# Notebook best practices
notebook_checklist = {
    "do": [
        "Start with a clear title and goal",
        "Use markdown cells to explain reasoning",
        "Show outputs inline (not just code)",
        "Include a requirements.txt",
        "Use relative paths for data",
        "Add a table of contents for long notebooks"
    ],
    "avoid": [
        "Long code cells without explanation",
        "Output cells that scroll for pages",
        "Hardcoded paths",
        "Unused imports",
        "Printing raw DataFrames",
        "Undefined variables"
    ]
}

# Docstring format (Google style)
def train_model(X_train, y_train, config):
    """Train a gradient boosting model with cross-validation.
    
    This function handles feature preprocessing, hyperparameter tuning,
    and model training in a single pipeline. It logs metrics to MLflow
    and saves the best model to the specified path.
    
    Args:
        X_train: Training features as a pandas DataFrame.
        y_train: Training labels as a numpy array.
        config: Configuration dictionary with keys:
            - n_estimators: Number of trees (default: 100)
            - max_depth: Maximum tree depth (default: 6)
            - learning_rate: Learning rate (default: 0.1)
    
    Returns:
        dict: Training results containing:
            - model: Trained model object
            - metrics: Cross-validation metrics
            - feature_importance: Feature importance scores
    
    Raises:
        ValueError: If X_train and y_train have different lengths.
        RuntimeError: If training fails to converge.
    
    Example:
        >>> config = {'n_estimators': 200, 'max_depth': 8}
        >>> results = train_model(X_train, y_train, config)
        >>> print(f"CV AUC: {results['metrics']['auc']:.4f}")
    """
    pass

# README template
readme_template = """
# Project Name

Brief description of what this project does and why.

## Quick Start

```bash
# Clone and setup
git clone <repo>
cd project
pip install -r requirements.txt

# Run
python main.py --config config.yaml

docker build -t ml-service .
docker run -p 8000:8000 ml-service

import matplotlib.pyplot as plt
import seaborn as sns

# Effective visualization checklist
viz_checklist = {
    "title": "Clear, descriptive title",
    "labels": "Axis labels with units",
    "legend": "Clear legend if needed",
    "colors": "Colorblind-friendly palette",
    "annotations": "Highlight key points",
    "source": "Data source attribution"
}

# Good visualization example
def plot_model_comparison(results):
    """Compare model performance with clear labels and context."""
    
    fig, axes = plt.subplots(1, 2, figsize=(12, 5))
    
    # Left: Metric comparison
    models = list(results.keys())
    aucs = [r['auc'] for r in results.values()]
    
    colors = ['#2ecc71' if a > 0.85 else '#3498db' for a in aucs]
    axes[0].barh(models, aucs, color=colors)
    axes[0].set_xlabel('AUC Score')
    axes[0].set_title('Model Performance Comparison')
    axes[0].axvline(x=0.85, color='red', linestyle='--', label='Target: 0.85')
    axes[0].legend()
    
    # Right: Latency vs Accuracy tradeoff
    latencies = [r['latency_ms'] for r in results.values()]
    axes[1].scatter(latencies, aucs, s=100)
    
    for i, model in enumerate(models):
        axes[1].annotate(model, (latencies[i], aucs[i]), 
                        textcoords="offset points", xytext=(5, 5))
    
    axes[1].set_xlabel('Inference Latency (ms)')
    axes[1].set_ylabel('AUC Score')
    axes[1].set_title('Accuracy vs Latency Tradeoff')
    
    plt.tight_layout()
    plt.savefig('model_comparison.png', dpi=150, bbox_inches='tight')
    plt.show()

# Mistake 1: Walls of code
bad_code_block = """
```python
import pandas as pd
import numpy as np
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import accuracy_score, roc_auc_score
import matplotlib.pyplot as plt
import seaborn as sns
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import Pipeline
# ... 50 more lines

from sklearn.ensemble import RandomForestClassifier

model = RandomForestClassifier(n_estimators=100, max_depth=6, random_state=42)
model.fit(X_train, y_train)