Claudio Silva

Claudio Silva

Institute Professor, New York University

CLAUDE.md

CLAUDE.md

Guidance for Claude Code when working with Claudio Silva’s academic website (Jekyll/GitHub Pages at ctsilva.github.io).

Quick Commands

# Setup
bundle install --path vendor/bundle

# Local server (http://127.0.0.1:4000/)
# --livereload auto-refreshes the browser on .md/_sass edits
bundle exec jekyll serve --livereload --incremental

# JavaScript build
npm run build:js

Mermaid Diagrams (for Quarto slides)

# Setup
npm install -g @mermaid-js/mermaid-cli

# Generate PNGs from .mmd files (script also exists in 2025-VisML-CSE/slides)
cd 2026-VisML-CDS/slides
python3 render_mermaid.py

# Manual generation
mmdc -i input.mmd -o output.png -w 800 -H 600 -t default -b transparent

Note: Always use PNG generation for RevealJS slides (inline mermaid doesn’t render reliably).

Site Structure

Jekyll Collections

  • _publications, _talks, _teaching, _portfolio, _posts, _pages
  • Publications in _pages/publications.md (chronological 2024→1996)
  • Media coverage in _pages/news.md (resilient linking with search fallbacks)

Key Files

  • _config.yml: Main configuration
  • markdown_generator/: Scripts for generating content from TSV/BibTeX

Course Websites

The repository includes complete course websites as subdirectories:

  • 2024-VisML-CDS: VisML (Visualization for Machine Learning) course website (Spring 2024, NYU CDS) - ARCHIVE
  • 2025-InfoVis-CSE: Information Visualization course website (Fall 2025, NYU Tandon) - CS-GY 6313 - ARCHIVE
  • 2025-VisML-CSE: Visualization for Machine Learning course website (Fall 2025, NYU Tandon) - CS-GY 9223 - ARCHIVE
  • 2026-InfoVis-CSE: Information Visualization (Fall 2026, NYU Tandon) - CS-GY 6313 B - ACTIVE
  • 2026-VisML-CDS: Visualization for Machine Learning (Fall 2026, NYU CDS) - DS-GA 3001 - ACTIVE

2026-InfoVis-CSE (CS-GY 6313 B)

  • Schedule: Fridays 11AM-1:30PM (Sept 4 - Dec 11), Jacobs Hall 6 MetroTech Room 315
  • No class: Nov 27 (Thanksgiving Recess). Fall Break (Oct 12) is a Monday and does not affect this course.
  • Sources copied from 2025; rendered .html was not copied — run quarto render per week before class.
  • Staff: TA (labs) and Grader are both Ryan Kim (returning from 2025; site https://www.rkim.dev, Observable handle @rk2546). Listed on home, syllabus, and week1-syllabus slides.
  • Instructor office hours: Fridays 1:30-2:30 PM, 370 Jay Street, Room 1153 (right after class).
  • Still stale in the lab decks: they point at Ryan’s 2025 resources — Observable notebooks under observablehq.com/@rk2546/2025-infovis-cse_week-N-lab and NYU Zoom rooms (nyu.zoom.us/j/92815268504, /2817596431, /98539408719), plus TA office hours listed as “Wednesdays 2-3PM, 8th floor common area @ 370 Jay”. Confirm with Ryan whether the notebooks get 2026 copies, whether the Zoom rooms and office hours are unchanged, and whether the separate “Grader TA” Zoom is still needed now that one person holds both roles. Then update the lab decks.

2026-VisML-CDS (DS-GA 3001)

  • Meeting pattern (corrected 2026-09-08 from the registrar listing): everything is on Tuesday. .001 Lecture Tuesdays 4:55-6:55 PM (120 min), then .002 Lab Tuesdays 7:10-8:00 PM (50 min) after a 15-min break, both in 60 Fifth Ave Room 150, cap 75 each. The lab is still its own registrar-scheduled section, so lab content does not consume lecture time — but it now lands at the end of a 3-hour evening rather than a day later, so a lab deck that overruns has no overnight reset. (An earlier version of this file recorded a Monday lecture as “confirmed”; that was wrong.)
  • Term: 14 Tuesdays, Sept 8 - Dec 8, with no cancellations — Labor Day (Sept 7) and Fall Break (Oct 12) are Mondays, Thanksgiving (Nov 26-27) is Thu-Fri. No Wednesday make-up is needed.
  • No final exam during exam week. Reading Day is Tue Dec 15 (no classes). Final report still due Dec 14 (term end), a week after the last class.
  • Instructor office hours: Fridays 1:30-2:30 PM, 370 Jay Street, Room 1153 (Brooklyn, not Washington Square).
  • Session leader (labs) & grader: Bhavya Matam (bm3792@nyu.edu).
  • No default project (decided 2026-10-12): every group proposes its own project; the proposal is the approval step. slides/default-project.qmd (Tile2Net) and projects/finalproject.md (StreetAware) are unlinked leftovers.
  • Project proposal timing (resolved 2026-10-12): discussion deck moved to Week 6 (Oct 13), proposal due Week 7 (Oct 20); project-discussion.qmd no longer carries 2025 TA names or dates.
  • Registrar deadlines (Fall 2026): add/drop full-semester Sept 15; withdraw & pass/fail Nov 25; Thanksgiving recess Thu-Fri Nov 26-27 (does not affect Mon/Tue sessions).
  • Sources copied from 2025-VisML-CSE; rendered .html was not copied — run quarto render per week before class.

  • Deck sizing against the real 120 min + 50 min split. Lecture ≈60 slides max; a hands-on 50-min lab is more like ~25. Tandon’s 150-min combined block is why several decks run long:

    WeekDateLecture deck(s)SlidesLab deckSlides
    1Sept 8course-intro + visualization-intro79 overvega-lite17 ok
    2Sept 15perception + color60 at limitperception-color11 ok
    3Sept 22model-assessment52 oknone — see below—
    4Sept 29white-box-models64 slightly overinterpretable-ml7 ok
    5Oct 6black-box-models + project-discussion82 overblack-box-explainability34 over
    6Oct 13clustering + project-discussion65 slightly over (6 clustering slides skippable)project-proposal (workshop; this is what ran)3 ok
    7Oct 20dimensionality-reduction36 (was 21; 14 new slides, all skippable)dimensionality-reduction (DR in practice; did not run Oct 13)48, about 25 core plus a backup block with jump links
    8Oct 27deep-learning32 ok (24 + 8 split-out text slides, laid out 2026-10-26)deep-learning (new; Playground + CNN Explainer, run by Claudio) + paper-writing (about 25 core min)12 + 39
    9Nov 3nlp35 oknlp-vis (rebuilt 2026-10-28: Embedding Projector, Transformer Explainer, LIME Colab lime_text.ipynb; chart QA survey is a backup block)23 core + 33 backup
    10Nov 10tda99 way overtda (Mapper, persistent homology)31 over
    11Nov 17Time Series — no deck yet———
    12Nov 24Interpretable ML & Fairness — no deck yet———

    Lecture problem is concentrated: Week 10 TDA (99 slides), then Weeks 5 and 1. Labs run long more often (six of nine) because 50 min is tight — but lab overflow is far less disruptive than lecture overflow.

  • Week 7 (Oct 20) deck and lab, built 2026-10-15. Lecture adds t-SNE equations, the crowding problem, live perplexity/seed/UMAP demos on sklearn digits, trustworthiness/continuity (a flat-with-perplexity result is shown on purpose), a PCA/t-SNE/UMAP table, a checklist, and a next-week slide; its Python cells need the uv venv and re-run on every render (about a minute). Speaker notes on the Distill/UMAP/TopoMap figure slides were rewritten against the figures (5 perplexities not 4, steps 10/20/60/120/1000, Fashion-MNIST not digits, TopoMap preserves components not holes). Lab labs/dimensionality-reduction loads sklearn’s 8x8 digits (64 features), not MNIST; there is a heads-up slide for that. Jump links need ids: [text](#lab-lle) (Quarto adds the /; writing #/lab-lle produces #//lab-lle). Known environment limits: the 3D plotly slides need WebGL and plotly’s CDN scripts (jquery, requirejs) and show “WebGL is not supported” in headless Chrome; the YouTube embed on the lab t-SNE slide errors (153) from file://.
  • File naming (renamed 2026-10-26): topic names, no week numbers. Lectures are slides/<topic>.qmd (deep-learning, nlp, tda, …). Labs are labs/<topic>/lab.qmd with a recap.md beside it (labs/deep-learning/, labs/nlp-vis/, labs/paper-writing/, …). Only schedule.md and home.md map weeks to files, so a schedule change never makes a filename lie. Old URLs for the 16 pages that had already been published are tiny redirect stubs at their old paths (e.g. slides/week2-intro.html); leave them. Not renamed on purpose: labs/week5/ (Colab links on the interpretable-ml lab point at that GitHub path), the loose labs/*.ipynb notebooks, and slides/default-project*. labs/archive/ holds 2025 recaps whose topic could not be verified. The Discord channel #week3-lab is still named in the perception-color lab.
  • Unresolved lab materials. Nine lab decks exist as .qmd and are scheduled. Still open:
    • Model Assessment lab (2026 Week 3, Sept 22). In 2025 this deck lived only in external Google Slides (linked from labs/model-assessment/recap.md), never in the repo — so it silently dropped out of the schedule. Currently listed on home/schedule as “materials in preparation”.
    • labs/guest-talk/ holds only a recap; the 2025 session was a guest presentation (Parikshit, external Google Slides). Not placed in the 2026 schedule — decide whether to repeat it.
    • Orphaned notebooks, not linked from any page: labs/Lab_white_box_vis1.ipynb, labs/Lab_white_box_vis2.ipynb (white-box, Week 4), labs/VisML_TDA_Lab.ipynb (TDA, Week 10).

2025-VisML-CSE (CS-GY 9223)

  • Schedule: Mondays 5-7:30PM (Sept 2-Dec 11), Room 473
  • Status: Week 11 (NLP and LLMs) COMPLETE ✅
  • Discord: https://discord.gg/dyHSFN65
  • Materials: week2-intro, week2-infovis, week2-lab, week4-model-assessment, week5-white-box, week6-black-box, week7-deep-learning, week11-nlp (file names as they were in 2025; the 2026 site uses topic names)

Course Material Enhancement Workflow

Trigger: “PLAN to do these updates for Week X”

  1. Rename generic figures: paperX.png → descriptive-name.png
  2. Standardize citations: Add footer citations with DOI links
  3. Add theoretical content: Include foundational slides with speaker notes
  4. Optimize for projector: Light code backgrounds, large centered videos
  5. Quality check: Render, test links, commit with descriptive messages
  6. Update docs: Mark week complete in CLAUDE.md

2025-InfoVis-CSE (CS-GY 6313)

  • Schedule: Fridays 11AM-1:30PM (Sept 5-Dec 13), Room 215
  • Status: Week 13 COMPLETE ✅ (Final presentations Dec 5 & 12)
  • Discord: https://discord.gg/sTEv3PnP
  • Materials: week1-syllabus, week2-data-transformation, week3-fundamental-graphs, week4-perception, week5-color, week6-group-projects, week7-interaction, week8-geovis-2d-maps, week9-urban-flows-interactivity, week10-urban-3d-simulation, week11-temporal-data, week12-clustering-dimreduction, week13-network-data
  • Resources: home.md, syllabus.md, schedule.md, slides/, labs/

Quarto Slide Creation

# Render slides
cd 2026-InfoVis-CSE/slides/
quarto render week1-syllabus.qmd

# Preview with live reload
quarto preview week1-syllabus.qmd

# Commit both HTML and _files/ directories
git add week1-syllabus.html week1-syllabus_files/
git commit -m "Add Week 1 slides"

The 2026 sites were copied without rendered output, so each deck needs a quarto render before its class. Decks with executable {python} cells (VisML model-assessment.qmd) need a Python with Jupyter, which the system Python lacks: uv venv /tmp/qpy && uv pip install --python /tmp/qpy/bin/python jupyter scikit-learn matplotlib seaborn pandas scipy plotly umap-learn, then QUARTO_PYTHON=/tmp/qpy/bin/python quarto render <deck>.qmd. *.html / *_files/ are globally ignored with a per-year negation in .gitignore — add new course years there or renders stay uncommitted.

Lab decks must be rendered --to revealjs. The VisML lab .qmds under labs/<topic>/lab.qmd (some, not all) declare both revealjs and html formats; a plain quarto render writes both to the same <name>.html and the html one wins, so the link on the site serves a scrolling page instead of slides (and speaker notes become div.notes, which reveal ignores). Their Python cells also need lime, shap and umap-learn on top of the venv above.

Template header:

---
title: "Title"
subtitle: "CS-GY 6313 - Fall 2026"
author: "Name"
format:
  revealjs:
    theme: [default, custom.scss]
    logo: figs/vida.jpg
    width: 1920
    height: 1080
---

Slide layouts: six canonical layouts (full-bleed, statement + figure, 40/60 text|figure, centred .nostretch, comparison, text-only) with the sizing rules that actually hold under this theme are in slide-templates/LAYOUTS.md; slide-templates/layout-templates.qmd renders one slide per layout. Pick the layout from the figure’s aspect ratio, not the amount of text. Two traps: width= alone does not disable auto-stretch (add .nostretch), and images inside .columns need height="760" (square/tall) or width="100%" (landscape).

Reference Management

PDF Naming: LastName_Year_Short_Title.pdf

Footer Citations:

:::footer
Author, A. (Year). [*Title*](../refs/Author_Year_Title.pdf). Venue.
:::

Venue Abbreviations: IEEE TVCG, CHI, ICML, AISTATS, ECML-PKDD