JAX PathwaysCheck your starting point
A matrix has three observations and two features. You need one mean per feature. Which axis should be reduced, and how many values remain?
Compare your reasoning
Reduce the observation axis, axis 0. Two feature means remain. Reducing axis 1 instead returns three observation means, which answers a different question.
If this is unfamiliar, work through From NumPy to jax.numpy before attempting the project.
Build in stages
1. Write the data contract
Annotate observation and feature axes; calculate one normalized row by hand. Compare a feature bias with an observation bias using unequal dimensions.
2. Make preprocessing safe to reuse
Pass parameters explicitly, preserve the caller’s arrays and check indexed accumulation separately from replacement. Carry shape, mask and valid-count checks into foundation-toolkit stage 2.
Try a changed condition
Make a prediction before opening the reasoning. Keep your answer and the evidence you would collect.
Normalize a matrix with a constant feature and one masked observation. What should the constant feature and the mean denominator do?
Compare an approach
Define the constant-feature policy before division, for example a unit fallback scale. Compute statistics only from valid observations; masking the final output alone leaves the fitted statistics contaminated. Reject a feature with no valid observations or declare an explicit policy.
Diagnose a likely failure
Symptom: Predictions look reasonable but adding a bias changes the wrong dimension.
Check next: Write the two shapes before broadcasting and test one row and one column independently.
Decide what is ready
Retain unequal-shape, constant-column and masked-row cases, a NumPy or hand reference, and proof that inputs were unchanged. Explain every axis in the output.
Where to put your project code
Write your code in: projects/arrays-audit/starter/model.py (copy projects/arrays-audit/starter/model.py first: cp projects/arrays-audit/starter/model.py projects/arrays-audit/starter/model.py).
Verify your implementation: python3 projects/arrays-audit/tests/check.py --stage all --impl projects/arrays-audit/starter/model.py
Run the staged project checks on your implementation in projects/arrays-audit/starter/model.py and record your transfer experiment results in your portfolio.
Open Connect arrays, transformations and explicit state
What still needs further work
Sparse and ragged inputs need additional contract-focused labs; this phase deliberately teaches fixed dense arrays first.