Skip to content

docs(readme): describe the redesigned workspace and verified features - #58

Open
alikahwaji wants to merge 1 commit into
mainfrom
docs/readme-redesign
Open

alikahwaji wants to merge 1 commit into
mainfrom
docs/readme-redesign

Conversation

@alikahwaji

Copy link
Copy Markdown
Contributor

Updates the README for the finished redesign and corrects claims that no longer matched the app. Every feature listed was checked against index.html and the code that wires it up.

What's new in the README

  • A workspace section covering the three columns, the Data → Model → Train → Evaluate workflow, and the Guided / Standard / Expert modes.
  • The feature list is regrouped by step (Data, Model, Training, Visualisation, Evaluation, Export, Learning & UX) and includes everything from redesign phases 1–4.
  • How to use is rewritten around the four steps.
  • The roadmap now records the codebase review and the four redesign phases, and lists candidates for what's next.

Corrections

  • Exists in code but has no UI, so it moved to "Implemented but not yet in the app":
    • ONNX export
    • Python code generation
    • boundary image export (PNG, SVG)
    • training-history CSV export
    • the live activation heatmap
    • early stopping, gradient clipping, warmup and cyclic learning-rate schedules
  • No implementation at all, so the claims are removed:
    • ROC/AUC
    • colour-scheme, opacity and contour sliders
    • browser notifications
    • session auto-save
    • ELI5 tooltips (archived)
    • model upload
  • Wrong details, now fixed:
    • mode names: Learn/Experiment/Advanced → Guided/Standard/Expert
    • classes: up to 10 → 2–4
    • presets: five → four
    • keyboard shortcuts: full list
    • TypeScript: 5.6 → 6
    • E2E browsers: CI runs Chromium only
    • the ML-backend example's train() returns TrainResult, not number

Docs only; no code changes.

🤖 Generated with Claude Code

Rewrites the README for the 2026 redesign: the three-column workspace,
the Data → Model → Train → Evaluate workflow, Guided/Standard/Expert
modes, and the new layer editor, Evaluate summary, overfitting-risk
meter, reproducible runs, training timeline and learning-rate finder.

Every feature claim was checked against the UI and wiring. Corrected:
- ONNX export, Python codegen, image export, history CSV and the
  activation heatmap exist as code but have no UI; moved to
  "Implemented but not yet in the app", with early stopping, gradient
  clipping and warmup/cyclic schedules (engine-only).
- Removed claims with no implementation: ROC/AUC, colour-scheme and
  opacity sliders, browser notifications, auto-save, ELI5 tooltips
  (archived), model upload.
- Mode names, classes (2–4, not 10), presets (four), shortcuts,
  TypeScript 6, CI browsers (Chromium in CI) and the ML-backend
  example's train() return type.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant