Springs [WIP]

[adamhaber]

Basic functionality for springs-regression animation.

Some todos:

1. Fix the division-by-zero error
2. ~~How to import all the functions already in better.clj? ~~
3. Prettify the animation (once it's working :-))
4. Add PCA animations (springs unconstrained on the rod's end)

[sritchie]

Very nice!! I'd love to see a screenshot, but I'll also run this myself!

[sritchie]

these are definitely generalized coordinates :)

If you wanted to decompose this further, you could make a "spring" lagrangian, that takes, for position, a sequence of spring-stretches only... and then compose that with a transformation that does what you do internally, maps the

(defn line->offsets [xs ys]
  (fn [[slope intercept]]
    (let [preds (map (linear-func slope intercept) xs)]
      (mapv - preds ys))))

Then

(defn L-regression [xs ys k]
  (compose (L-springs k) (F->C (line->offsets xs ys))

should work! Just an idea...

[adamhaber]

Implemented this, ran into a different problem, see comment below.

[sritchie]

this is very cool. If you take my idea from above, about composing, I think you could use the exact same transformation to get positions for all of the points.

[adamhaber]

I still can't get the graphics to work; the things I'm curious about (but can't experiment with at the moment) - if the drawing function only takes in the state - the intercept and the slope - how can I draw the xs and ys?

[sritchie]

I want to see it!!

[adamhaber]

Lol me too :-)

[adamhaber]

Thanks for your comments Sam!

I've implemented

(defn L-regression-composed [xs ys k]
  (compose (L-springs k) (F->C (line->offsets xs ys))))

But now, when I call

(let [L (L-regression-composed x y 1)
      initial-state (up 1
                        (up 1 1)
                        1)
      built (build L initial-state)]
  (q/defsketch regression
    :title "Regression line"
    :size [500 500]
    :setup (:setup built)
    :update (:update built)
    :draw (draw-line (:xform built))
    :features [:keep-on-top]
    :middleware [m/fun-mode m/navigation-2d]))

I get:

Execution error (IllegalArgumentException) at sicmutils.generic/bin+ (generic.cljc:177).
No method in multimethod 'add' for dispatch value: [:sicmutils.structure/up :sicmutils.expression/numerical-expression]

When I use the original L-regression instead of the composed version (shouldn't they be identical?), I get the previous Division by zero error. Any ideas?

[adamhaber]

Any idea why Calva/cljfmt (my Clojure VSCode extension) forces the double spacing here? Same for the "fake data" comment below, but most comments don't have this.

[sritchie]

strange, I do not!!

[sritchie]

Hey, I'll load up your code in the next couple of days and give it a run myself to see what's up here!