https://github.com/harmonydata/experiments

Experiments with Harmony
https://github.com/harmonydata/experiments

auc evaluation harmony harmonydata large-language-model llm natural-language-processing nlp roc

Last synced: 14 days ago
JSON representation

Experiments with Harmony

• Host: GitHub
• URL: https://github.com/harmonydata/experiments
• Owner: harmonydata
• License: mit
• Created: 2023-07-13T09:11:49.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2024-04-10T09:26:54.000Z (9 months ago)
• Last Synced: 2024-04-10T10:36:33.030Z (9 months ago)
• Topics: auc, evaluation, harmony, harmonydata, large-language-model, llm, natural-language-processing, nlp, roc
• Language: Jupyter Notebook
• Homepage: https://harmonydata.ac.uk
• Size: 1.1 MB
• Stars: 0
• Watchers: 1
• Forks: 1
• Open Issues: 0
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
# experiments

Experiments with Harmony

## Factor loading

Can Harmony reproduce factor loadings from the SDQ? Attempting to reproduce factor loadings from the 5-factor model from [Karlsson et al](https://doi.org/10.1371/journal.pone.0265481).

## Kufungisisa

How does Harmony handle cultural concepts of distress such as *kufungisisa* (thinking too much) in Shona?

I had to use a Shona transformer model for this. 

Using [Shona symptom questionnaire for the detection of depression and anxiety](https://depts.washington.edu/edgh/zw/hit/web/project-resources/shona_symptom_questionnaire.pdf) and running in Google Colab.

## Multilingual evaluation

Harmony can now handle 8 languages (at least):

* Portuguese, French, German, Spanish, Russian, Chinese, Hebrew