fix(fit_line): keep the sample weights non-negative

[ducha-aiki]

What does this pull request do?

nbs/fit_line.ipynb builds its weights as 1 - p2_noise.mean(), where p2_noise = torch.rand_like(p2) * std and std = 1.2. The mean of two U(0, 1.2) draws is above 1 with probability ≈ 5.6%, so with 50 unseeded points at least one weight is negative ≈ 94% of the time. Since kornia/kornia#5108, fit_line rejects negative weights (ValueCheckError: fit_line requires non-negative weights), and the tutorials job of kornia's scheduled CPU tests has failed on this notebook in most runs since 2026-10-03, for example https://github.com/kornia/kornia/actions/runs/37397633192/job/112057348727.

The weight is now 1 - p2_noise.mean() / std. It lies in [0, 1] and keeps the original intent: noisier points get less weight.

What did you check?

I executed the notebook's code cells (plotting cell excluded) against current kornia main, with seeds 0–199:

before: 185 / 200 seeds raise "fit_line requires non-negative weights"
after:    0 / 200

The notebook still parses as JSON. Only that one source line changes.

Posted on behalf of @ducha-aiki by Claude (Opus 5.5).

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB