Simple Machine Learning Code Tutorial for Beginners with Sklearn Scikit-Learn

Summary

This tutorial demonstrates a beginner-friendly machine-learning workflow with Scikit-learn, using California housing data to predict house values. It covers preparing data, training and evaluating models, trying ways to improve them, and saving a trained model for reuse.

Main Ideas and Workflow

1. Set Up the Project

Create and activate a Conda environment with Python 3.12. Install Scikit-learn and Jupyter, create a project folder, and launch JupyterLab.

2. Load and Inspect the Data

Fetch the California housing dataset from Scikit-learn. The input columns—such as income, house age, rooms, population, and location—are the features, conventionally named X. The home values are the target, conventionally named y.

Inspect the feature names and sample records to understand what the inputs and target represent.

3. Split Data into Training and Test Sets

Use train_test_split to reserve 20% of the data for testing and use the remaining 80% for training. Set a random_state to make the shuffled split repeatable. Without it, the split—and therefore the results—can change between runs.

4. Train and Evaluate a Baseline Model

Start with linear regression and fit it to the training features and targets. Then predict targets for the test features, without giving the model the test targets.

Compare predictions with the actual test targets using the R² score. The presenter reports a baseline score of about 0.60. R² is not classification accuracy; it indicates how well the model accounts for variation in the target compared with a baseline.

5. Try Polynomial Features

Expand the original eight features using polynomial features. The example produces 45 features, including squared features and combinations of pairs of features. The presenter reports that this raises the R² score to about 0.66.

The goal is to give the model additional patterns in the data that it can use to make predictions.

6. Compare Algorithms

Test linear regression alongside random forest and gradient-boosting regressors. The example finds that the tree-based methods perform better than the initial linear-regression baseline, illustrating why it is useful to experiment rather than assume one algorithm will work best.

To speed up random forest, set n_jobs=-1 so it can use all available CPU cores. The presenter also switches to histogram-based gradient boosting, reporting that it runs faster and improves the score.

7. Tune Model Hyperparameters

After selecting a promising model, try different settings for max_iter (the number of boosting iterations) and learning_rate. Use nested loops to test combinations and compare their R² scores.

The example reports a best result around 0.84. The narration first identifies 300 iterations with a learning rate of 0.05 as the best combination, but later saves a model configured with 350 iterations and the same learning rate. This is an inconsistency in the subtitles.

The broader lesson is to evaluate parameter combinations systematically rather than rely on a single default configuration.

8. Save and Reload the Model

Use Joblib to save the fitted model to a file. Load it again with joblib.load and make predictions or calculate R² using the reloaded model.

The tutorial reports that the loaded model reproduces the same score, showing how a trained model can be reused without training it again.

Speakers and Sources Featured

• Speaker: One unnamed presenter or narrator from the Python Simplified channel.
• Data source: Scikit-learn’s California housing dataset.
• Tools and libraries: Scikit-learn, JupyterLab, Conda, and Joblib.