Measure whether the 3x3 plus-sign classifier works

Translate the claim into assertions or a tiny eval harness. The metric to protect is: accuracy on a fixed four-image set; feature activations printed. Always record the denominator (how many cases) beside any rate. A percentage without a denominator is marketing, not measurement.

tests=[([[0,1,0],[1,1,1],[0,1,0]],'plus'),([[1,1,1],[0,0,0],[0,0,0]],'other')]
def pred(x): return 'plus' if sum(x[r][c] for r,c in [(0,1),(1,0),(1,1),(1,2),(2,1)])>=4 else 'other'
assert all(pred(x)==y for x,y in tests); print('2/2')

Expected evidence: fragile pixel classifier. A passing assertion proves only the behavior it names; broader usefulness still needs the chapter’s full limits.

Be explicit: beating the baseline (pixel-sum threshold accuracy on the same four images) on this fixture does not prove behavior under overfitting to one plus orientation, or leaking test grids into template design. Label observations separately from conclusions so the next page inherits honest evidence about the 3x3 plus-sign classifier.

Compute accuracy on a fixed four-image set; feature activations printed with the denominator written beside the rate every time. For this chapter, the evaluation set is intentionally tiny; that is allowed only if you say so in the evidence. Compare against pixel-sum threshold accuracy on the same four images before celebrating.

Add one negative case aimed at overfitting to one plus orientation, or leaking test grids into template design. A suite with only happy cases cannot protect the 3x3 plus-sign classifier when the characteristic failure appears in review.

If a check is expensive or flaky, shrink it until it is deterministic on plus, blank, and distractor 3×3 grids. Flaky green builds teach the team to ignore gates. Record what this page does not prove so security-ops and mastery-ship inherit honest limits.

At the evaluation stage for computer-sees, the job is narrower than finishing a product demo. You are creating one progressive evidence piece about plus, blank, and distractor 3×3 grids that later pages inherit without redefining success. Keep that fixture small enough to inspect by hand, keep outputs copy-pasteable as text, and refuse to narrate this baseline as if it were a production SLA: pixel-sum threshold accuracy on the same four images.

For this page specifically, success looks like metrics with explicit denominators and a negative case while still centering the user decision to classify tiny binary images as plus vs not-plus with inspectable features. If you cannot point to a file, command, or assertion that proves that for the 3x3 plus-sign classifier, stay on this page instead of advancing.

Glossary: computer vision · Glossary: convolutional neural network

Previous · Next

Return to the computer sees scenario and restate what this chapter proved on page validation.mdx. Name one metric, one ownership rule, and one regression test you will keep. Explain how this page connects to the previous page without repeating earlier paragraphs. If you cannot name a falsifier, the chapter is still a story rather than a controlled practice. Write the falsifier as an observable event with a threshold.

Return to the computer sees scenario and restate what this chapter proved on page validation.mdx. Name one metric, one ownership rule, and one regression test you will keep. Explain how this page connects to the previous page without repeating earlier paragraphs. If you cannot name a falsifier, the chapter is still a story rather than a controlled practice. Write the falsifier as an observable event with a threshold.