How To Calculate R2 Score Examples Numpy Python

Calculate the coefficient of determination (R²) for your regression model

Comprehensive Guide: How to Calculate R² Score with NumPy in Python

The R-squared (R²) score is a fundamental statistical measure that evaluates how well a regression model explains the variance in the dependent variable. This guide provides a complete walkthrough of calculating R² scores using NumPy in Python, with practical examples and theoretical explanations.

1. Understanding R² Score Fundamentals

The coefficient of determination, denoted as R², represents the proportion of variance in the dependent variable that’s predictable from the independent variable(s). Its mathematical definition is:

R² = 1 – (SS_res / SS_tot) where: – SS_res = Sum of squares of residuals (actual – predicted) – SS_tot = Total sum of squares (actual – mean(actual))

Key properties of R²:

• Ranges from 0 to 1 (though negative values are possible with poor models)
• 1 indicates perfect prediction
• 0 indicates the model performs no better than a horizontal line
• Negative values suggest the model performs worse than a horizontal line

2. Manual Calculation Example

Let’s calculate R² manually for this dataset:

Observation	Actual (Y)	Predicted (Ŷ)
1	3	2.8
2	5	5.1
3	7	6.9
4	9	9.2

Step-by-step calculation:

1. Calculate mean of actual values: (3+5+7+9)/4 = 6
2. Calculate SS_tot: (3-6)² + (5-6)² + (7-6)² + (9-6)² = 20
3. Calculate SS_res: (3-2.8)² + (5-5.1)² + (7-6.9)² + (9-9.2)² = 0.14
4. R² = 1 – (0.14/20) = 0.993

3. NumPy Implementation

Python’s NumPy library provides efficient array operations for R² calculation:

import numpy as np def calculate_r2(y_true, y_pred): # Calculate mean of actual values y_mean = np.mean(y_true) # Calculate total sum of squares ss_tot = np.sum((y_true – y_mean) ** 2) # Calculate sum of squares of residuals ss_res = np.sum((y_true – y_pred) ** 2) # Calculate R² score r2 = 1 – (ss_res / ss_tot) return r2 # Example usage y_actual = np.array([3, 5, 7, 9]) y_predicted = np.array([2.8, 5.1, 6.9, 9.2]) r2_score = calculate_r2(y_actual, y_predicted) print(f”R² Score: {r2_score:.4f}”)

4. Using scikit-learn for R² Calculation

While this guide focuses on NumPy, scikit-learn provides a convenient function:

from sklearn.metrics import r2_score # Using the same data score = r2_score(y_actual, y_predicted) print(f”R² Score (sklearn): {score:.4f}”)

5. Practical Applications and Interpretation

R² scores find applications across various domains:

Domain	Typical R² Range	Interpretation
Physics	0.95-1.00	Highly deterministic relationships
Economics	0.50-0.80	Moderate predictive power
Social Sciences	0.20-0.50	Weak but meaningful relationships
Marketing	0.10-0.30	Low predictive power

Important considerations when interpreting R²:

• R² always increases with more predictors (adjusted R² accounts for this)
• Not suitable for comparing models with different dependent variables
• Should be used alongside other metrics like RMSE, MAE
• Sensitive to outliers in the data

6. Common Pitfalls and Solutions

Avoid these mistakes when working with R² scores:

1. Overfitting: High R² on training data but poor performance on test data.
   Solution: Use cross-validation and holdout sets.
2. Ignoring baseline: Not comparing against simple models.
   Solution: Always compare with a naive baseline (e.g., mean prediction).
3. Non-linear relationships: R² assumes linear relationships.
   Solution: Use polynomial features or non-linear models when appropriate.
4. Small sample sizes: R² can be misleading with few observations.
   Solution: Use adjusted R² or collect more data.

7. Advanced Topics

Adjusted R²: Penalizes adding non-contributing predictors:

n = len(y_true) # number of observations p = y_true.shape[1] if len(y_true.shape) > 1 else 1 # number of predictors adjusted_r2 = 1 – (1-r2)*(n-1)/(n-p-1)

Partial R²: Measures contribution of individual predictors in multiple regression.

R² for Non-linear Models: Can be calculated similarly but interpretation differs.

8. Real-world Example: Housing Price Prediction

Consider predicting house prices (in $1000s) based on square footage:

House	Square Feet	Actual Price	Predicted Price
1	1500	300	295
2	2000	350	360
3	2500	420	425
4	3000	480	490
5	3500	550	555

Calculating R² for this dataset:

y_actual = np.array([300, 350, 420, 480, 550]) y_pred = np.array([295, 360, 425, 490, 555]) r2 = calculate_r2(y_actual, y_pred) print(f”Housing R²: {r2:.3f}”) # Output: 0.998

This exceptionally high R² (0.998) indicates the linear model explains 99.8% of price variance based on square footage alone.

Authoritative Resources

For deeper understanding, consult these academic resources:

• NIST Engineering Statistics Handbook – R² Section (National Institute of Standards and Technology)
• BYU Statistics Department – R² Interpretation Guide (Brigham Young University)
• NIH Publication on Regression Metrics in Biomedical Research (National Institutes of Health)

Frequently Asked Questions

Q: Can R² be negative?

A: Yes, when your model performs worse than a horizontal line (predicting the mean). This typically indicates a very poor model or incorrect implementation.

Q: What’s the difference between R² and adjusted R²?

A: Adjusted R² accounts for the number of predictors in the model. It penalizes adding non-contributing variables, while regular R² always increases with more predictors.

Q: How is R² related to correlation coefficient?

A: In simple linear regression, R² equals the square of the Pearson correlation coefficient (r) between actual and predicted values.

Q: What’s a good R² value?

A: This depends entirely on your domain. In physics, you might expect 0.99+, while in social sciences 0.2 might be considered good. Always compare against domain baselines.

Q: Can I use R² for classification problems?

A: No, R² is specifically for regression problems. For classification, use metrics like accuracy, precision, recall, or AUC-ROC.