A Symbolic Regression tool to search for symbolic expressions which minimize residuals to an objective, written in Clojure.
Warning: This is experimental software, with an unstable API. Expect breaking changes prior to a major release.
- Draw an objective function, select a built-in dataset, or provide a CSV
- Select number of points, iterations
- See progress
- Chart of best fitting function
- Chart of loss over time
- A selectable text version of the function
Run the application from binary, download a release JAR and see the Usage section below.
This project is not yet available in a JAR repository (though I'd use Clojars if enough requests).
Or, you can run the app from source, using Clojure development tools (leiningen) in this project:
$ lein deps/repl/run/uberjar...
Use the application via GUI or in the terminal.
There are 4 ways to run the application. See next section for details about the options you can provide when running the app.
Requirements: Java
$ java -jar closyr-0.1.0-SNAPSHOT-standalone.jar ...options here...
Requirements: Java, Leiningen
$ lein run
Or headless (just in the terminal), where you can specify options and input data:
$ lein run -t -p 25 -i 5 -x 0,1,2,3,4,5,6 -y 1,2,30,4,5,6,10
Which is the same as:
$ lein run --headless --population 1000 --iterations 200 --xs 0,1,2,3,4,5,6 --ys 1,2,30,4,5,6,10
Requirements: Java, Leiningen (I will provide a deps file if enough interest)
(require '[closyr.symbolic-regression :as symreg])
(symreg/run-app-with-gui)
Requirements: Java, Leiningen
$ lein uberjar
$ java -jar target/uberjar/closyr-0.1.0-standalone.jar
You can also provide the same command-line options to java command, like:
$ java -jar target/uberjar/closyr-0.1.0-standalone.jar -t -p 25 -i 5 -x 0,1,2,3,4,5,6 -y 1,2,30,4,5,6,10
| Short, Long Option | Required? | Example | Default | Description |
|---|---|---|---|---|
-t,--headless |
no | -t |
false |
run without GUI, in terminal only |
-c,--use-flamechart |
no | -c |
false |
run with flamecharts, run then visit http://localhost:54321/flames.svg |
-p,--population |
no | 100 |
20 |
size of population which will evolve; the number of functions we create and modify |
-i,--iterations |
no | 50 |
10 |
number of iterations to run for |
-l,--max-leafs |
no | 40 |
40 |
max number of AST tree leafs in candidate functions |
-x,--xs |
no, unless ys |
1,3,4 |
random | the xs for the points in the dataset to fit against; if provided, must also provide ys and be the same count |
-y,--ys |
no, unless xs |
2,4,8 |
random | the ys for the points in the dataset to fit against; if provided, must also provide xs and be the same count |
-f,--infile |
no | f.csv |
A CSV file. Contains either 2 columns without titles in first row, or has columns x and y to be used as objective dataset |
On successful application start, you can start an run a search. You might see something like this when done:
An example of using the GUI:
$ lein test
Coverage looks like this if you run lein cloverage:
- We use Genetic Algorithms to allow candidate functions of best fit to compete.
- They compete on a computed score on the input data, and we use the sum of residuals to generate the score.
- Evolution consists of mutation and crossover.
- Mutations act on the function's AST and modify branches, leafs, or the whole tree. Operations like
+0.1,*x,/Sin(x)are applied to functions. - Crossovers combine two functions' ASTs at a random point, and combine using various operators like
+, and*.
- CLI options accept a CSV file (GUI already supports this)
- More tests
- Use something like ProGuard to shrink the JAR for releases
- A different frontend. The current Java Swing frontend does the job, but it's not easy to maintain and it's hard to make look better.
- Can this be a follow-up to this issue, asking for a symbolic regression tool on the JVM? axkr/symja_android_library#850
- When I created this and my other symbolic regression tools, I didn't know about the formal field of symbolic regression. I've since found some great libraries that I should review and apply the lessons to this project: https://github.com/MilesCranmer/PySR
- This project would not have been possible without the symbolic math library
Symja(https://github.com/axkr/symja_android_library), which has been great to use and I consider it to be like aMathJs(https://github.com/josdejong/mathjs) on the JVM.
I accept PRs, please open an issue to discuss beforehand to make sure we are in alignment. Thanks!
Copyright © 2024 Octavian Geagla
This program and the accompanying materials are made available under the terms of the Eclipse Public License 2.0 which is available at http://www.eclipse.org/legal/epl-2.0.
This Source Code may also be made available under the following Secondary Licenses when the conditions for such availability set forth in the Eclipse Public License, v. 2.0 are satisfied: GNU General Public License as published by the Free Software Foundation, either version 2 of the License, or (at your option) any later version, with the GNU Classpath Exception which is available at https://www.gnu.org/software/classpath/license.html.
________/\\\\\\\\\__/\\\___________________/\\\\\__________/\\\\\\\\\\\____/\\\________/\\\____/\\\\\\\\\_____
_____/\\\////////__\/\\\_________________/\\\///\\\______/\\\/////////\\\_\///\\\____/\\\/___/\\\///////\\\___
___/\\\/___________\/\\\_______________/\\\/__\///\\\___\//\\\______\///____\///\\\/\\\/____\/\\\_____\/\\\___
__/\\\_____________\/\\\______________/\\\______\//\\\___\////\\\_____________\///\\\/______\/\\\\\\\\\\\/____
_\/\\\_____________\/\\\_____________\/\\\_______\/\\\______\////\\\____________\/\\\_______\/\\\//////\\\____
_\//\\\____________\/\\\_____________\//\\\______/\\\__________\////\\\_________\/\\\_______\/\\\____\//\\\___
__\///\\\__________\/\\\______________\///\\\__/\\\_____/\\\______\//\\\________\/\\\_______\/\\\_____\//\\\__
____\////\\\\\\\\\_\/\\\\\\\\\\\\\\\____\///\\\\\/_____\///\\\\\\\\\\\/_________\/\\\_______\/\\\______\//\\\_
_______\/////////__\///////////////_______\/////_________\///////////___________\///________\///________\///__