Loss — pick the better model

Background

A loss is the ruler that scores how wrong a model is: lower is better. So comparing two models on the same labels is just comparing their losses. Here we use mean squared error,

$$L_{\text{MSE}} = \frac{1}{n}\sum_{i=1}^{n}(y_i - \hat{y}_i)^2.$$

Problem statement

Implement better_model(y, preds_a, preds_b) returning which model's predictions have the lower MSE against y.

Input

• y — array-like of true targets.
• preds_a — array-like, model A's predictions (same length as y).
• preds_b — array-like, model B's predictions.

Output

Returns an int: 0 if model A has the lower (or equal) MSE, 1 if model B is strictly lower.

Examples

Example 1

Input:  y = [1, 0, 1], preds_a = [0.9, 0.1, 0.8], preds_b = [0.2, 0.9, 0.1]
Output: 0

Explanation: A's MSE $\approx 0.02$ vs B's $\approx 0.75$ — A is far better.

Example 2 — B wins

Input:  y = [0, 0], preds_a = [1, 1], preds_b = [0.1, 0.1]
Output: 1

Explanation: B is much closer to the zeros.

Constraints

• Compute each model's MSE and compare.
• On a tie, return 0 (prefer A).
• Return a Python int (0/1).

Notes

• "Training = lower loss" is the whole game; this is the comparison the optimiser makes implicitly at every step.