https://github.com/Golem-Unlimited/gudot
Linear regression using FHE gMorph library
https://github.com/Golem-Unlimited/gudot
Last synced: 8 months ago
JSON representation
Linear regression using FHE gMorph library
- Host: GitHub
- URL: https://github.com/Golem-Unlimited/gudot
- Owner: golemfactory
- License: gpl-3.0
- Created: 2019-09-16T14:47:06.000Z (over 6 years ago)
- Default Branch: master
- Last Pushed: 2023-01-20T22:27:13.000Z (over 3 years ago)
- Last Synced: 2024-10-01T04:27:56.263Z (over 1 year ago)
- Language: Rust
- Size: 138 KB
- Stars: 2
- Watchers: 5
- Forks: 3
- Open Issues: 7
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# gudot
This is a showcase for feasibility of gMorph/FHE on Golem.
It computes linear regression using FHE [gMorph] library.
The whole software is experimental and is provided for demonstration purposes only.
Please note that we make no claims about security of the encryption scheme.
It is included in Golem's workshop on DevCon 5.
[gMorph]: https://github.com/golemfactory/gmorph
## Requirements
* emsdk - installation instructions can be found
[here](https://emscripten.org/docs/getting_started/downloads.html)
* gwasm-runner - you can either build it from source as described
[here](https://github.com/golemfactory/gwasm-runner), or download
a precompiled binary from [here](https://github.com/golemfactory/gwasm-runner/releases)
* freetype library - on Ubuntu install package `libfreetype6-dev`
## Building
Assuming you've got the prerequisites satisfied, in order to build the project, simply run
```
cargo build --release
```
## Running
The project involves executing several steps in sequence: some steps are done on the host OS (so your machine),
while the main step is performed using gWasm (run on `gwasm-runner`).
For illustration purposes, every step is executed by a separate command:
```
$ target/release/gudot generate
Writing input.json
$ target/release/gudot encrypt
Reading input.json
Writing enc_input.json
Writing keys.json
$ gwasm-runner target/release/gudot.wasm -- --subtasks=N
work dir: .../split
Reading enc_input.json
Writing enc_output.json
$ target/release/gudot decrypt
Reading keys.json
Reading enc_output.json
Writing output.json
$ target/release/gudot regress
Reading output.json
Reading input.json
Fitted model: y = -2.6898656093724984x + 93773.13902413758
Writing regress.json
$ target/release/gudot plot
Reading input.json
Reading regress.json
Writing plot.png
```
## Running on Golem network
```
$ gwasm-runner --backend=Brass target/release/gudot.wasm -- --subtasks=N
```
When running on the Golem network, you may want to adjust some parameters specific to Brass Golem
(e.g. the path to your Golem's data directory). To do so, you'll need to create a configuration
file for the runner. You can find more information on this in the [gwasm-runner README].
[gwasm-runner README]: https://github.com/golemfactory/gwasm-runner/blob/master/README.md#running-on-the-golem-network
## Components
* `generate` --- creates `input.json` containing sample data in the format `[[x1,...,xn][y1,...,yn]]`
(this step is optional - you can provide your own `input.json`);
* `encrypt` --- generates a secret encryption key (`keys.json`), and encrypts `input.json` yielding `enc_input.json`;
* `gudot.wasm` --- the main gWasm binary, meant to be run with `gwasm-runner`; it splits the encrypted data;
sends parts to providers and combines partial results returned by the providers
* `decrypt` --- decrypts the received results, producing `output.json` (reads `keys.json` and `enc_output.json`);
* `regress` --- calculates parameters `(slope, intercept)` of the fitted linear model, producing `regress.json`
* `plot` --- plots the regression line along with the input points, producing `plot.png`.