Skip to content
CompStats PlaygroundMAST90083 · 2023 S2

Showcase · workflows and screenshots

A guided tour.

Three short recordings of the main workflows, each step captioned on screen, followed by screenshots of every feature. They were made by a scripted browser tour against this site, so anyone with the repository can re-record them and get the same seeds and the same numbers.

3 walkthroughs captions and transcripts 17 screenshots, light, dark and mobile

Workflow walkthroughs

Watch the main journeys

Videos are silent. Each step is captioned on screen, the player also offers the captions as a text track, and the numbered steps beside each video are its transcript, with the time each step starts.

Walkthrough 1 · 1:03 · /regularisation

Shrinkage

Drag log λ and watch ridge and lasso paths, the cross-validation curve and test error respond.

Open Regularisation

Steps and transcript

  1. 10:00Ridge regression on 263 Hitters players. λ starts at the 2023 cross-validated λmin, 300.9.
  2. 20:05Drag log λ: every ridge coefficient shrinks smoothly towards zero, but none reaches it.
  3. 30:19Switch to the lasso: its ℓ₁ penalty sets coefficients to exactly zero, one at a time.
  4. 40:32Jump to the lasso's λmin, 4.468: 13 of 19 predictors are kept, as in the 2023 PDF.
  5. 50:38The 10-fold CV curve (±1 SE band) and test MSE against OLS follow the same slider.
  6. 60:562026 upgrade: a paired bootstrap over the 132 test players. Every difference's 95% interval includes zero.

Walkthrough 2 · 0:52 · /model-selection

Pick the order

Flip the 2023 data bug on and off, then run the AR Monte Carlo at n = 15 and n = 100 and compare the criteria with intervals.

Open Model selection

Steps and transcript

  1. 10:00As submitted in 2023, Q2.4 built its 'AR' series from the Hitters salaries left over from Q1.
  2. 20:06Corrected: series simulated from the AR models themselves. The selected orders change.
  3. 30:15Monte Carlo in a Web Worker: true model M2, an AR(2), with n = 15 and 1,000 replicates, seed 10.
  4. 40:20At n = 15 every criterion overfits. The Wilson 95% intervals for the true order sit near zero.
  5. 50:29Same seed, n = 100.
  6. 60:35Now the BIC-like IC₃ finds p = 2 most often. Paired McNemar tests compare criteria on the same replicates.
  7. 70:44Optional AI explanations use your own key. Without one nothing is sent, and the site works fully.

Walkthrough 3 · 0:58 · /svm, /bootstrap

Kernels & bootstrap

Switch SVM kernels and watch the decision regions retrain, then run the bootstrap bias animation and the repeated-experiment check.

Open SVM Open Bootstrap

Steps and transcript

  1. 10:00Three simulated Gaussian classes and a LIBSVM port, retrained live. Linear kernel, C = 10, as in Q1.2.
  2. 20:05Radial kernel: the regions bend around the classes. Drag cost C to trade margin violations against fit.
  3. 30:18Polynomial kernel. Shading is the predicted class and outlined points are support vectors.
  4. 40:25The bootstrap: B resampled estimates of θ pile up. Their mean minus the plug-in value is the bias estimate.
  5. 50:31Raise B and draw new samples: the bias estimate and the corrected estimator update.
  6. 60:43Is the corrected estimator unbiased? Repeated experiments check b₁ and b₂ against the theory.

Screenshots

Every feature at a glance

Desktop shots are 1440 × 900. Select one to open a larger view; the arrow keys move between them.

On a phone (390 px)

How these were made

  • A Playwright script drives Google Chrome through each journey at a human pace, adds the caption banner and a visible cursor, and records the screen at 1280 × 800. The same script is an end-to-end test: it checks the numbers on screen (λmin = 300.9, 13 of 19 predictors, 67 support vectors) as it goes.
  • Every seed is the site's default or set by the script, so a re-recording shows the same results. The recordings are trimmed and converted with ffmpeg; the captions above come from the same step list.
  • No AI key was entered. The AI settings dialog is shown empty, and nothing was sent to an AI provider.