bug fix

[Mike Tokic (mitokic)]

This pull request updates the Croston model handling to ensure that historical actuals are preserved in backtests, CSV outputs, and model averages, while continuing to clamp negative targets only during Croston training. It also adds comprehensive tests to validate this behavior and updates documentation accordingly.

Bug Fixes and Improvements for Croston Model:

• Croston now preserves historical actuals in backtests, downstream CSV outputs, and model averages; negative targets are still clamped to zero for Croston training only.
• The clamp_negative_target function now saves the original Target values in a Target_Original column (if not already present) before clamping, ensuring that assessment actuals remain unchanged and only training data is affected.

Testing Enhancements:

• Added tests to verify that Croston clamping preserves assessment actuals, does not alter training data outside of training splits, and that CSV/model outputs retain historical actuals. Also checks that the clamping is a no-op for other models and when no negatives exist. [1] [2]

Documentation and Metadata Updates:

• Updated the package version to 0.7.0.9013 in DESCRIPTION and NEWS.md to reflect these changes. [1] [2]

[Copilot]

🔵 Needs a closer look

Validation remains unverified; devtools::test(filter = "clamp_negative_target") and devtools::check() need confirmation before approval.

0 open findings

What changed in this PR

Preserves Croston historical actuals throughout FinnTS backtests and downstream outputs while retaining non-negative training targets.

Changes:

• Saves original targets before Croston clamping.
• Adds regression coverage for assessment actuals, CSV outputs, and model averages.
• Updates release notes and package version to 0.7.0.9013.

File	Description
tests/​testthat/​test-clamp_negative_target.R	Tests actual preservation and downstream outputs.
R/​train_models.R	Preserves original targets before clamping.
NEWS.md	Documents the Croston correction.
DESCRIPTION	Advances the development version.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.