scarp / by example
Menu

CHAPTER 06 · 11 LESSONS

06 · Foundations for AI-oriented programs

Scarp has early numerical primitives useful for AI-related calculations. These exercises do not claim full AI frameworks, autodiff, GPU acceleration, model training systems or agent tooling.

English guide ↗ हिन्दी अध्याय ↗ Coding lab for this chapter ↗

YOUR CHAPTER LEARNING PATH · सीखने का क्रम

Read. Check. Build.

Three linked steps using real Scarp source and verified output.

  1. Read the lessons · पाठ पढ़ें

    Work through all 11 verified examples. Save optional local lesson progress as you learn; this does not certify code execution.

    Start with Matrix Inference ↗

    Lesson tracking is optional and only available when browser storage works.

  2. Check understanding · समझ जाँचें

    Try 3 bilingual questions linked directly to their original source and expected output. No quiz score is stored.

    Review all three questions here · यहाँ अभ्यास करें ↓

  3. Write Scarp · कोड लिखें

    Edit the real starter locally, compare output with the compiler-verified goal, and use the hint or tested solution when needed.

    Open Choose a class as a one-hot vector lab ↗

Lesson completion is voluntary local tracking. Passing a quiz or lab is not automatically recorded or claimed by this website. The Scarp compiler runs locally, not in your browser.

Lessons in this chapter

01 Matrix Inference Matrix/tile construction and matrix-vector multiplicationQuick check · प्रश्न 02 Classification Gradient One-hot target and explicit categorical-loss gradientQuick check · प्रश्न 03 Classification Loss Scalar categorical loss and selecting maximum score indexQuick check · प्रश्न 04 Score Selection Deterministic argmax/argmin and extremum values 05 Matrix Targets tile, matrix-vector scores, argmax and onehot 06 Matrix Forward Typed tile input, tileadd, tilescale, matvec, bias, softmax and contracts 07 Log Space logsumexp and logsoftmax without exponential overflow 08 Coefficient Fit User-written squared-error loss and gradient steps on small vectors 09 Linear Transform matmul composition of a 2×3 and a 3×2 tile 10 Class Score Ties matvec, argmax first-match behavior and maxval 11 Matrix Probabilities Equal-score softmax and categorical-loss tolerance check
Practice this chapter · इस अध्याय का अभ्यास करें 3 bilingual, compiler-evidenced questions

Answer all three questions using the checked-in lessons. Feedback runs only in your browser; there is no stored score, automatic lesson completion or Scarp execution. Answers remain readable without JavaScript.

Question 1 of 3 · Matrix Inference · View source and recorded output ↗

SOURCE-BACKED SELF-CHECK · स्रोत-आधारित अभ्यास

Quick checkpoint / त्वरित अभ्यास

What output does matvec(weights, features) produce?matvec(weights, features) का आउटपुट क्या है?

Reveal the source-backed answer / सही उत्तर देखें

[4.0, 1.0] — With rows [1.0, 2.0] and [-1.0, 3.0], multiplying by [2.0, 1.0] gives 4.0 and 1.0.[4.0, 1.0] — मैट्रिक्स की पंक्तियाँ [1.0, 2.0] और [-1.0, 3.0] हैं। इन्हें [2.0, 1.0] से गुणा करने पर 4.0 और 1.0 मिलते हैं।

This question checks understanding of a recorded compiler output. It does not run Scarp in the browser. यह प्रश्न सत्यापित आउटपुट पर आधारित है; ब्राउज़र में Scarp नहीं चलता।

Question 2 of 3 · Classification Gradient · View source and recorded output ↗

SOURCE-BACKED SELF-CHECK · स्रोत-आधारित अभ्यास

Quick checkpoint / त्वरित अभ्यास

What gradient does lossgrad([0.0, 0.0], 1) print?lossgrad([0.0, 0.0], 1) कौन-सा gradient प्रिंट करता है?

Reveal the source-backed answer / सही उत्तर देखें

[0.5, -0.5] — Equal logits imply probabilities of 0.5 for both classes. The loss gradient subtracts the one-hot target for class 1, producing [0.5, -0.5].[0.5, -0.5] — समान logits के लिए दोनों वर्गों की संभावना 0.5 होती है। class 1 के one-hot target को घटाने पर gradient [0.5, -0.5] मिलता है।

This question checks understanding of a recorded compiler output. It does not run Scarp in the browser. यह प्रश्न सत्यापित आउटपुट पर आधारित है; ब्राउज़र में Scarp नहीं चलता।

Question 3 of 3 · Classification Loss · View source and recorded output ↗

SOURCE-BACKED SELF-CHECK · स्रोत-आधारित अभ्यास

Quick checkpoint / त्वरित अभ्यास

Which index does argmax([1.0, 4.0, 2.0]) print?argmax([1.0, 4.0, 2.0]) कौन-सा index प्रिंट करता है?

Reveal the source-backed answer / सही उत्तर देखें

1 — The maximum value is 4.0, which is at zero-based index 1. The fixture prints 1.1 — सबसे बड़ा मान 4.0 है, जो शून्य-आधारित index 1 पर है। इसलिए आउटपुट 1 है.

This question checks understanding of a recorded compiler output. It does not run Scarp in the browser. यह प्रश्न सत्यापित आउटपुट पर आधारित है; ब्राउज़र में Scarp नहीं चलता।

All examples are validated with Scarp v0.0.6-alpha.3. Each lesson has its own code and expected-output page.