MACHINE LEARNING 101

MACHINE LEARNING 101

A GUIDE TO TRAINING YOUR FIRST MACHINE LEARNING MODEL

GET STARTED

OBJECTIVES

Welcome to this interactive guide on setting up your first machine learning modelThis guide will walk you through each practical step from data preparation to deployment By the end, you'll have created a functioning machine learning model

TOOLS NEEDED

PYTHON

JUPYTER

SCIKIT-LEARN

PANDASMATPLOTLIB FLASK

NEXT PAGE

INDEX

GUIDE

ASSESSMENT

NEXT STEPS

GUIDE

DATA PREPARATION

MODEL SELECTION

FEATURE SELECTION

EVALUATION METRICS

MODEL IMPROVEMENT

TRAINING PROCESS

PREDICTIONS WITH MODEL

MODEL DEPLOYMENT

ASSESSMENT

INDEX

PREPARING YOUR DATA

NEXT PAGE

Download the sample housing dataset provided Load the data using pandas with this code snippet Examine your data structure: data.head() Check for missing values: data.isnull().sum() Handle missing values with appropriate techniques Convert categorical features to numeric

DOWNLOAD DATA

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

INDEX

FEATURE SELECTION + ENGINEERING

NEXT PAGE

Examine correlation between features and target Visualize key relationships Create new meaningful features Select the most relevant features

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

CODE SNIPPET 4

INDEX

CHOOSING YOUR MODEL

NEXT PAGE

For our housing price prediction (regression task), we'll compare:

• Linear Regression

• Random Forest

• Gradient Boosting

Import and set up models Split data into training and testing sets

LR

RF

XGB

CODE SNIPPET 1

CODE SNIPPET 2

INDEX

TRAINING your model

NEXT PAGE

Train each model on your prepared data Monitor training process (for supported models) Check training completion and model details

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

INDEX

EVALUATION METRICS

NEXT PAGE

Make predictions on test data Calculate regression metrics Visualize predictions vs. actual values

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

INDEX

IMPROVING YOUR MODEL

NEXT PAGE

Prevent overfitting with cross-validation Tune hyperparameters to improve performance Retrain with optimal parameters

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

INDEX

MAKING PREDICTIONS

NEXT PAGE

Prepare a function to make new predictions Use your model for predictions Save your model for future use

CODE SNIPPET 1

CODE SNIPPET 2

CODE SNIPPET 3

INDEX

MODEL DEPLOYMENT

ASSESSMENT

CODE SNIPPET 1

Create a simple Flask API to serve predictions Test your API with a sample request Monitor your model in production

CODE SNIPPET 2

CODE SNIPPET 3

ASSESSMENT

LET'S GO

NEXT STEPS

Learning Resources:

• Online courses: Fast.ai, Coursera Machine Learning Specialization
• Websites: Kaggle competitions, Towards Data Science blog
• Documentation: scikit-learn, pandas, and matplotlib references

Next Project Ideas:

• Classification project: Customer churn prediction
• Time series project: Stock price forecasting
• NLP project: Sentiment analysis of product reviews

Learning Path:

• Progression from regression to classification problems
• Suggested skill development roadmap
• More advanced housing price models to try next

from sklearn.linear_model import LinearRegression from sklearn.ensemble import RandomForestRegressor, GradientBoostingRegressor # Initialize models linear_model = LinearRegression() forest_model = RandomForestRegressor(n_estimators=100, random_state=42) boost_model = GradientBoostingRegressor(n_estimators=100, random_state=42)

from flask import Flask, request, jsonifyimport joblib app = Flask(__name__) # Load the model model = joblib.load('housing_price_model.pkl') @app.route('/predict', methods=['POST']) def predict(): # Get data from request data = request.get_json() # Make prediction price = predict_house_price( model=model, bedrooms=data['bedrooms'], bathrooms=data['bathrooms'], square_feet=data['square_feet'], year_built=data['year_built'], neighborhood=data['neighborhood'] ) # Return prediction return jsonify({'predicted_price': price}) if __name__ == '__main__': app.run(debug=True)

import requestsimport json # Test data test_house = { 'bedrooms': 4, 'bathrooms': 2.5, 'square_feet': 2200, 'year_built': 2010, 'neighborhood': 'downtown' } # Send request to API response = requests.post('http://localhost:5000/predict', json=test_house) # Print response result = response.json() print(f"API response: ${result['predicted_price']:.2f}")

plt.figure(figsize=(10, 6))plt.scatter(y_test, forest_pred, alpha=0.5) plt.plot([min(y_test), max(y_test)], [min(y_test), max(y_test)], 'r--') plt.xlabel("Actual Prices") plt.ylabel("Predicted Prices") plt.title("Random Forest: Predicted vs. Actual Housing Prices") plt.show()

from sklearn.model_selection import cross_val_score # 5-fold cross-validation cv_scores = cross_val_score(forest_model, X, y, cv=5, scoring='neg_mean_absolute_error') # Convert negative MAE to positive and print results mae_scores = -cv_scores print(f"Cross-validation MAE scores: {mae_scores}") print(f"Average MAE: ${mae_scores.mean():.2f}")

# Predict price for a specific houseprice = predict_house_price( model=final_model, bedrooms=3, bathrooms=2, square_feet=1800, year_built=2005, neighborhood='suburb' ) print(f"Predicted house price: ${price:.2f}")

# Add logging to your APIimport logging logging.basicConfig(filename='model_predictions.log', level=logging.INFO) @app.route('/predict', methods=['POST']) def predict(): # Get data from request data = request.get_json() # Make prediction price = predict_house_price( model=model, bedrooms=data['bedrooms'], bathrooms=data['bathrooms'], square_feet=data['square_feet'], year_built=data['year_built'], neighborhood=data['neighborhood'] ) # Log prediction logging.info(f"Input: {data}, Prediction: {price}") # Return prediction return jsonify({'predicted_price': price})

from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_scoreimport numpy as np # Calculate metrics for each model def evaluate_model(model_name, predictions, actual): mae = mean_absolute_error(actual, predictions) rmse = np.sqrt(mean_squared_error(actual, predictions)) r2 = r2_score(actual, predictions) print(f"{model_name} Performance:") print(f"MAE: ${mae:.2f}") print(f"RMSE: ${rmse:.2f}") print(f"R² Score: {r2:.4f}") print("-" * 30) evaluate_model("Linear Regression", linear_pred, y_test) evaluate_model("Random Forest", forest_pred, y_test) evaluate_model("Gradient Boosting", boost_pred, y_test)

# For Gradient Boosting with early stoppingfrom sklearn.model_selection import train_test_split X_train, X_val, y_train, y_val = train_test_split(X_train, y_train, test_size=0.2) boost_model = GradientBoostingRegressor(n_esti mators=500, validation_fraction=0.2, n_iter_no_change=5, random_state=42) boost_model.fit(X_train, y_train)

from sklearn.model_selection import GridSearchCV # Define parameter grid param_grid = { 'n_estimators': [50, 100, 200], 'max_depth': [None, 10, 20, 30], 'min_samples_split': [2, 5, 10] } # Set up grid search grid_search = GridSearchCV( RandomForestRegressor(random_state=42), param_grid=param_grid, cv=3, scoring='neg_mean_absolute_error', return_train_score=True ) # Perform search grid_search.fit(X, y) # Get best parameters print(f"Best parameters: {grid_search.best_params_}") print(f"Best MAE: ${-grid_search.best_score_:.2f}")

def predict_house_price(model, bedrooms, bathrooms, square_feet, year_built, neighborhood): # Create a DataFrame with the same structure as training data input_data = pd.DataFrame({ 'bedrooms': [bedrooms], 'bathrooms': [bathrooms], 'square_feet': [square_feet], 'property_age': [2023 - year_built], 'price_per_sqft': [0] # Will be calculated }) # Add neighborhood one-hot encoding for n in ['downtown', 'suburb', 'rural']: input_data[f'neighborhood_{n}'] = [1 if neighborhood == n else 0] # Calculate price per sqft using average from our dataset # In a real scenario, this would be calculated after prediction avg_price = 250000 # Example average price input_data['price_per_sqft'] = avg_price / input_data['square_feet'] # Select only the features used in training input_features = input_data[features] # Make prediction predicted_price = model.predict(input_features)[0] return predicted_price