https://github.com/zeionara/relentness
Knowledge graph models manipulator which allows to perform various evaluations of provided embedders
https://github.com/zeionara/relentness
cross-validation graph-embeddings graph-embeddings-evaluation openke rotate swift transe wikidata
Last synced: 3 months ago
JSON representation
Knowledge graph models manipulator which allows to perform various evaluations of provided embedders
- Host: GitHub
- URL: https://github.com/zeionara/relentness
- Owner: zeionara
- License: apache-2.0
- Created: 2021-10-12T20:20:12.000Z (over 3 years ago)
- Default Branch: master
- Last Pushed: 2023-02-22T22:19:59.000Z (over 2 years ago)
- Last Synced: 2025-01-02T07:46:16.503Z (5 months ago)
- Topics: cross-validation, graph-embeddings, graph-embeddings-evaluation, openke, rotate, swift, transe, wikidata
- Language: Swift
- Homepage:
- Size: 534 KB
- Stars: 0
- Watchers: 3
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# relentness
Knowledge graph models manipulator which allows to perform various evaluations of provided embedders. Currently there is only an incomplete support for fetching **wikidata** triples and saving them in the format acceptable for the OpenKE toolkit (see examples of the generated datasets in the folder `Assets/Demo`).
# Setup
For being able to run python-based code you need to create a new environment and install the project dependencies:
```sh
conda env create --file environment.yml
```
# Usage
Currently on hyperparameter search in supported, which can be initiated using the following command which configures 2 concurrent workers to process cross-validation splits for a given dataset and `Assets/Log/default.txt` as a log destination file path:
```sh
swift run relentness hsearch -e grapex-latest 17 19 -n 2 -l default
```
To generate a sample from wikidata using default params, execute the following command:
```sh
swift build && ./.build/debug/relentness Assets/Demo
```
The generated files will be saved in the directory `Assets/Demo`. To list all available option for the `sample` subcommand, please, use the following call:
```sh
./.build/debug/relentness sample --help
```
To test a model on a given dataset with transe model:
```sh
python -m relentness test ./Assets/Corpora/Demo/0000 -m transe
```
Aside from `transe` there is a `complex` model supported as well at the moment which requires a similar call.
# Integrations
## Google sheets
The applications allows to export results of models comparison into google spreadsheet with nice formatting which automatically highlights all the main points.
## Telegram API
It is possible to interact with the application via telegram bot. Upon startup you are able to provide any access token and fine-tune the app to send a wide range of notifications, the most interesting of which is the message that informs you about the need of updating OAuth2 access token to a third-party service. On the screen below an example of a short chat with telegram bot is provided.
