Basic simplex optimiser #1

richfitz · 2022-05-09T14:55:24Z

No description provided.

codecov · 2022-05-09T14:57:31Z

Codecov Report

❗ No coverage uploaded for pull request base (main@a5da4a7). Click here to learn what that means.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main       #1   +/-   ##
=======================================
  Coverage        ?   96.06%           
=======================================
  Files           ?        3           
  Lines           ?      127           
  Branches        ?       25           
=======================================
  Hits            ?      122           
  Misses          ?        5           
  Partials        ?        0

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update a5da4a7...6a9f56e. Read the comment docs.

EmmaLRussell

Looks great to me, not having any huge insight into the algorithm. Just a couple of questions about the implications for wodin.

There seem to be a couple of packages on npm doing roughly this already - but they don't offer the ability to hook into each iteration?

EmmaLRussell · 2022-05-17T08:46:10Z

README.md

+while (true) {
+  if (opt.step()) {
+    break;
+  }
+}


Suggested change

while (true) {

if (opt.step()) {

break;

}

}

while (!opt.step()) {

// do something

}

Can we get data from opt in the while loop to peek at the state during the optimisation? Oh, looking at the example below, looks like you need to make an additional simplex call on opt to get the state out? So the global function simplex runs the whole optimise, but the class method simplex just gives you the current state? That's a little confusing, maybe that could be called something different since the class is already called Simplex?

Agree, I'll rename the free function fitSimplex or something like that perhaps

EmmaLRussell · 2022-05-17T08:55:05Z

README.md

+* The `step()` method advances the algorithm one step, which will take one or two evaluations of the target function and may or may not find a better point than our current best. It returns `true` if we have converged.
+* The `result()` method returns the object described above.
+
+Sometimes, it is useful to get additional information out of the target function and store it alongside the solution. For example, if fitting a model to data you might end up with a sum-of-squares error used for target fitting, but the model values themselves are of interest. For linear regression (which you would fit with a different packge!) we might do this by returning a closure that could be evaluated at any `x` position:


So the target function can either just return a numerical value, or an object with interface {value: number, data: any} where data could be any arbitrary data, or a closure or whatever you like? For odin, will we use this? The current parameter values being used will be given in location? And then we could run the model as we usually do with those values to plot the current solution, or would that be returned in data to save having to run it again?

So the target function can either just return a numerical value, or an object with interface {value: number, data: any} where data could be any arbitrary data, or a closure or whatever you like?

Yes, exactly that. I did look at requiring the more complex type but it's pretty ugly. Working on an example at the moment that uses the full return type and it seems to work ok.

So the target function can either just return a numerical value, or an object with interface {value: number, data: any} where data could be any arbitrary data, or a closure or whatever you like? For odin, will we use this? The current parameter values being used will be given in location? And then we could run the model as we usually do with those values to plot the current solution, or would that be returned in data to save having to run it again?

Basically, yes - there will be no need to run the model with those parameters again

Weird, it swallowed my comment. Which was, I didn't quite follow the comment in parentheses - you're saying that you've used linear regression as a toy example but wouldn't actually recommend using this package for linear regression? And I was initially confused about the expected scope of x and y in theta below, maybe include the x and y arrays in the example to make that clear?

yes, one should definitely not use this for linear regression! I've updated the comment and added arrays to make that less confusing

perfect, thanks

EmmaLRussell · 2022-05-17T09:09:05Z

example/example.js

+}
+
+function showSimplex(id, obj) {
+    var points = obj.simplex().map(el => el.location);


simplex returns an array, not just the current state? Oh, because it includes all the points in the simplex, not just the best one? So what should we be showing on the fit plot at that point? the solution for the first location in the simplex is always the best so show that?

yes, first is always the best. But it's not that exciting to look at and I was curious how the simplex itself evolves. You can see it conforming to the shape of the surface, even though it does not explicitly have gradient information, which is cool

EmmaLRussell · 2022-05-17T09:10:36Z

package.json

+    ],
+    "scripts": {
+        "build": "tsc",
+        "test": "nyc mocha",


We usually use jest for js testing, but don't mind giving mocha a try!

I am 100% happy to move to what we use currently - could you point me at something similar to crib off? I can move it over to ts tests too at the same time if that's what everything else uses

Maybe check out vue-charts, that's doesn't have too much unnecessary guff in jest.config.js: https://github.com/reside-ic/vue-charts (You won't need jest-canvas-mock)

moved these over, not massively different in the end, which was nice

EmmaLRussell · 2022-05-17T09:15:51Z

src/simplex.ts

+}
+
+export function simplex(target: TargetFn, location: number[],
+                        control: Partial<SimplexControlParam> = {},


It would be good to mention control and maxIterations params in the README.

EmmaLRussell · 2022-05-17T09:34:01Z

test/test-control.js

@@ -0,0 +1,30 @@
+"use strict";


Tests could be ts too if you like

this is done now

Co-authored-by: EmmaLRussell <44669576+EmmaLRussell@users.noreply.github.com>

README.md

Co-authored-by: EmmaLRussell <44669576+EmmaLRussell@users.noreply.github.com>

richfitz added 7 commits May 9, 2022 15:55

Start importing code from js prototype

c3ac64b

Start testing optimiser

c927a1a

Update test

3918e05

Test initialisation

f3fc0ba

Small tidyup

b3df5ca

Simplify and expand tests

7335a79

Add gha

fdddc93

richfitz and others added 9 commits May 13, 2022 15:04

Export Simplex class

48215bf

Allow storing extra data

e737c40

Add polymorphic interface

4abe3d9

Remove debugging

e885425

Fix lint

add6975

Add support for extracting simplex

89e88d1

Rename throughout inteface

bf4ae10

Add example

1356375

Basic docs

0c362e9

richfitz marked this pull request as ready for review May 16, 2022 14:27

richfitz added 2 commits May 16, 2022 15:45

Rename package

3d56092

Rename, again

9c2e1e8

richfitz requested review from EmmaLRussell and hillalex May 16, 2022 15:50

EmmaLRussell reviewed May 17, 2022

View reviewed changes

richfitz and others added 7 commits May 17, 2022 17:54

Rename high-level interface

803e720

Expand docs

6fcd1e0

Port tests to ts/jest

462b390

Fix compiler options

c63da4e

Use esnext modules

e5dadb4

Configure webpack

4f654bd

Update to use webpack

5ac2cd9

richfitz and others added 2 commits May 18, 2022 09:56

Update README.md

0ae5d8f

Co-authored-by: EmmaLRussell <44669576+EmmaLRussell@users.noreply.github.com>

Fix webpack export

7693fd0

richfitz requested a review from EmmaLRussell May 18, 2022 09:58

hillalex approved these changes May 18, 2022

View reviewed changes

Make example and comment clearer

639454e

EmmaLRussell reviewed May 18, 2022

View reviewed changes

README.md Outdated Show resolved Hide resolved

Update README.md

6a9f56e

Co-authored-by: EmmaLRussell <44669576+EmmaLRussell@users.noreply.github.com>

EmmaLRussell approved these changes May 18, 2022

View reviewed changes

richfitz merged commit 6172b3e into main May 18, 2022

richfitz mentioned this pull request Jun 27, 2022

Add docs for dfoptim #2

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Basic simplex optimiser #1

Basic simplex optimiser #1

richfitz commented May 9, 2022

codecov bot commented May 9, 2022 •

edited

EmmaLRussell left a comment

EmmaLRussell May 17, 2022

EmmaLRussell May 17, 2022

richfitz May 17, 2022

EmmaLRussell May 17, 2022

richfitz May 17, 2022

hillalex May 18, 2022

richfitz May 18, 2022

hillalex May 18, 2022

EmmaLRussell May 17, 2022

richfitz May 17, 2022

EmmaLRussell May 17, 2022

richfitz May 17, 2022

EmmaLRussell May 17, 2022 •

edited

richfitz May 18, 2022

EmmaLRussell May 17, 2022

EmmaLRussell May 17, 2022

richfitz May 18, 2022

Basic simplex optimiser #1

Basic simplex optimiser #1

Conversation

richfitz commented May 9, 2022

codecov bot commented May 9, 2022 • edited

Codecov Report

EmmaLRussell left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

EmmaLRussell May 17, 2022 • edited

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented May 9, 2022 •

edited

EmmaLRussell May 17, 2022 •

edited