sourcelibs/Learn/demos/blobs.xtl

1⍝!/usr/bin/env xetal 2⍝│ Learn's four models on one set of points: three blobs of 20 (around 3⍝│ 0 0, 6 6 and 0 6, a fixed ripple instead of chance). k-means finds 4⍝│ them without their labels; kNN and logistic regression learn them 5⍝│ from their labels; PCA finds the direction they spread along most. 6⍝│ Run with "just demo-lib Learn blobs". 7 8"ml:" u̲se< "Learn" 9 10i ← o̲ffsets 60 11blob ← 1 + i d̲iv 20 12cx ← (1 + blob = 2) s̲elect 0.0 6.0 13cy ← (1 + blob = 1) s̲elect 6.0 0.0 14X ← o̲\ 2 60 r̲eshape (cx + 0.8 × s̲in 1.3 × f̲loat i) c̲at cy + 0.8 × c̲os 2.1 × f̲loat i 15⍝│ k-means: the three centers found, and which cluster each point is in. 16C ← (3, 10) ᵐˡk̲means X 17ᵘr̲ound ← { a → (f̲loat f̲loor 0.5 + 100.0 × a) ÷ 100.0 } 18ᵘr̲ound C 19C ᵐˡa̲ssign X 20⍝│ kNN (5 neighbors) and logistic regression on four new points. 21Q ← 4 2 r̲eshape 1.0 1.0 5.0 5.0 0.5 5.0 3.0 3.0 22(5, X, blob) ᵐˡk̲nn Q 23W ← (200, 0.5) ᵐˡl̲ogistic (X, blob) 24W ᵐˡp̲redict Q 25⍝│ PCA: the direction of most spread, and the share of the variance on it. 26(mu, V, ev) ← 2 ᵐˡp̲ca X 27ᵘr̲ound r̲avel 1 t̲ake₂ V 28ᵘr̲ound (1 s̲elect ev) ÷ '+ r̲/ ev