save_component carries the component's imports with it

[stefan-jansen]

The defect

A component is validated by re-executing its saved source in an empty namespace with only numpy and pandas in scope. That is the right design - it is what makes a component survive a cold Colab session - but it means a student who writes

from sklearn.linear_model import Ridge      # an import cell

and then uses Ridge inside their component had their correct model recorded as not conformant.

Both remedies the error message suggests fail on an imported class, reproduced before the fix:

Remedy	Result
also=[Ridge]	ValueError: Your model_linear refers to 'MultiOutputMixin' - it tries to inline sklearn's own source
include={'Ridge': Ridge}	SyntaxError - it writes the class's repr into the file

This was the one place the conformance machinery was brittle in a way a student experiences as us being wrong, and it fires on the most ordinary thing they can do.

The fix

save_component works out the imports from the names the component actually mentions and writes them into the saved file as imports. also carries a function, include carries a value, and an import now travels as what it is.

Name resolution walks up to the shallowest ancestor module that still exports the object, so the saved file records the path the student typed:

# Saved by ml4t-coursework. This is your own code, exactly as you wrote it.
import numpy as np
import pandas as pd

from sklearn.linear_model import Ridge

class LinearModel:
    ...

rather than sklearn.linear_model._ridge, which is what the class reports as its own module and which the library is free to rename.

The name collection is deliberately not scope-aware. It over-collects - a local variable's name lands in the candidate set too - and the caller-namespace lookup is what filters, because a local name does not resolve to an importable object. Precise scoping would be more code for the same result.

What is unchanged

The genuine missing-symbol case is still refused, and there is a test asserting it: a value or helper defined in another cell does not resolve to an importable object, so it falls through to the existing message naming the symbol.

Verification

• 124 passed, 1 skipped. Three new tests: an imported class travels; a module imported under an alias travels; a value from another cell is still refused.
• ruff check src tests clean.
• The original failing reproduction now saves conformant, with a reference delta of 1.2e-18 against the shipped ridge.

🤖 Generated with Claude Code

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.