Introduction to ML Quick Reference

Quick lookup guide for ML concepts, algorithm selection, and workflow.

Table of Contents

• Algorithm Selection Decision Tree
• ML Workflow Checklist
• Problem Type Identification
• Quick Terminology Lookup
• Common Patterns and Anti-patterns

Algorithm Selection Decision Tree

ML Workflow Checklist

Problem Definition

• Understand business problem
• Define success metrics
• Identify constraints (time, resources)
• Determine data requirements

Data Collection

• Identify data sources
• Collect initial dataset
• Check data quality
• Document data sources

Data Preparation

• Handle missing values
• Remove duplicates
• Handle outliers
• Encode categorical variables
• Scale/normalize features
• Split data (train/val/test)

Feature Engineering

• Create new features
• Select important features
• Remove irrelevant features
• Handle feature interactions

Model Training

• Choose baseline model
• Train model
• Evaluate on validation set
• Tune hyperparameters
• Try different algorithms
• Compare models

Evaluation

• Evaluate on test set
• Check for overfitting
• Analyze errors
• Calculate metrics
• Compare with baseline

Deployment

• Save model
• Create API/interface
• Deploy to production
• Set up monitoring
• Document model

Monitoring

• Track predictions
• Monitor performance
• Check for drift
• Plan retraining

Problem Type Identification

Regression Problems

Characteristics:

• Target is continuous (numbers)
• Examples: Price, temperature, sales, age

Key Indicators:

• "Predict the value of..."
• "Estimate how much..."
• "Forecast..."

Metrics:

• MSE, RMSE, MAE, R²

Classification Problems

Characteristics:

• Target is categorical (classes)
• Examples: Spam/Not spam, Cat/Dog/Bird, High/Medium/Low

Key Indicators:

• "Classify as..."
• "Identify which..."
• "Predict the category..."

Metrics:

• Accuracy, Precision, Recall, F1, ROC-AUC

Clustering Problems

Characteristics:

• No labels available
• Find groups in data
• Examples: Customer segments, document topics

Key Indicators:

• "Group similar..."
• "Find patterns in..."
• "Segment..."

Metrics:

• Silhouette score, Inertia, Davies-Bouldin index

Quick Terminology Lookup

Data Terms

Model Terms

Evaluation Terms

Common Patterns and Anti-patterns

Good Patterns

# Pattern 1: Proper train/test split
from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

# Pattern 2: Scale features
from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
X_train_scaled = scaler.fit_transform(X_train)
X_test_scaled = scaler.transform(X_test)  # Don't fit on test!

# Pattern 3: Cross-validation
from sklearn.model_selection import cross_val_score

scores = cross_val_score(model, X_train, y_train, cv=5)
print(f"Mean CV score: {scores.mean():.3f}")

# Pattern 4: Evaluate on test set only once
model.fit(X_train, y_train)
test_score = model.score(X_test, y_test)

Anti-patterns

# Anti-pattern 1: Data leakage
# DON'T: Use future information
df['target'] = df['feature'].shift(-1)  # Using future data!

# Anti-pattern 2: Fitting on test data
# DON'T: Fit scaler on test data
scaler.fit(X_test)  # Wrong!

# Anti-pattern 3: Overfitting
# DON'T: Use too complex model without validation
model = ComplexModel(max_depth=1000)  # Will overfit!

# Anti-pattern 4: Testing multiple times
# DON'T: Evaluate on test set multiple times
for i in range(10):
    model = train_model()
    score = model.score(X_test, y_test)  # Wrong! Test set used multiple times

Quick Algorithm Reference

Regression Algorithms

Classification Algorithms

Clustering Algorithms

Common Mistakes to Avoid

1. Data Leakage: Using future information to predict past
2. Overfitting: Model too complex, memorizes training data
3. Underfitting: Model too simple, can't learn patterns
4. Wrong Metrics: Using accuracy for imbalanced data
5. No Validation, not using validation set
6. Test Set Contamination: Using test set multiple times
7. Ignoring Baseline, not comparing with simple baseline
8. No Documentation, not documenting experiments

Quick Tips

1. Start Simple: Begin with simple models (linear/logistic regression)
2. Use Baseline: Always compare with a simple baseline
3. Validate Properly: Use cross-validation or hold-out validation
4. Check for Overfitting: Compare train vs validation performance
5. Use Appropriate Metrics: Choose metrics based on problem type
6. Document Everything: Keep track of experiments and results
7. Iterate: ML is iterative, improve based on results

Try next: Use this page while coding. Open the main guide when a line here is unclear.