utility-function-extractor/packages/utility-tools/README.md

63 lines
2.2 KiB
Markdown
Raw Normal View History

2022-06-19 00:14:07 +00:00
# Utility Tools
2022-06-19 00:14:07 +00:00
This package contains a series of utilities to work with the utility functions produced by [this utility function extractor](https://utility-function-extractor.quantifieduncertainty.org/openphil-2018-ai-risk).
2022-06-19 00:14:07 +00:00
## Built with
2022-06-19 00:14:07 +00:00
- [Squiggle](https://www.squiggle-language.com/)
- [Nodejs](https://nodejs.org/)
- Plain js
2022-06-19 00:14:07 +00:00
## Getting started
### Installation
```sh
yarn add utility-tools
```
then in your file:
```js
import { mergeSort, findPaths, aggregatePaths } from "utility-tools";
```
### Usage
You can find an example how to use and concatenate these functions in `/src/example.js`, as well as an example of the input format needed in the `input` folder.
## Interface
### Merge sort (`mergeSort`)
Given a list of elements and a list of utilitity comparisons, sort the list. If there are not enough comparisons to implement the merge sort algorithm, return one of the missing comparisons.
_Gotcha_: The list of elements has to be the same list, and in the same order, as that produced when initially doing the comparisons. This is because the merge-sort algorithm depends on the initial order of the list.
### Find Paths (`findPaths`)
Given an (ordered) list of elements and a list of utility comparisons, find all possible monotonous paths from each element to each other. A monotonous path is a path that is either all increasing or all decreasing, relative to the ordering given.
_Note_: Some elements will have many more paths than others.
2022-06-20 00:50:53 +00:00
_Note_: The `findPaths.js` file has a few un-used functions which should make it easier to understand the code.
2022-06-19 00:14:07 +00:00
### Aggregate paths (`aggregatePaths`)
Given a list of path, aggregate them to finally produce an estimate of the relative utility of each element.
There are two ways of doing this:
- 1. Aggregate the means (expected values) for each path.
- This method is fast
- But has the disadvantage the expected value aggregation is tricky, particularly if one of the elements is positive and the other one negative (because then one can't)
- 2. Aggregate the distributions given for each path.
## Roadmap
I don't have any additions planned for this repository.
## Contact
Feel free to shoot me any questions at `nuno.semperelh@protonmail.com`