https://github.com/piesposito/transformers-low-code-experiments
Low-code pre-built pipelines for experiments with huggingface/transformers for Data Scientists in a rush.
https://github.com/piesposito/transformers-low-code-experiments
deep-learning machine-learning nlp pytorch transformer
Last synced: 2 months ago
JSON representation
Low-code pre-built pipelines for experiments with huggingface/transformers for Data Scientists in a rush.
- Host: GitHub
- URL: https://github.com/piesposito/transformers-low-code-experiments
- Owner: piEsposito
- License: apache-2.0
- Created: 2020-10-14T14:20:34.000Z (almost 5 years ago)
- Default Branch: main
- Last Pushed: 2020-10-14T18:04:01.000Z (almost 5 years ago)
- Last Synced: 2025-04-23T21:33:32.026Z (6 months ago)
- Topics: deep-learning, machine-learning, nlp, pytorch, transformer
- Language: Python
- Homepage:
- Size: 226 KB
- Stars: 16
- Watchers: 1
- Forks: 3
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Transformers for Data Scientists in a rush
Low-code pre-built pipelines for experiments with huggingface/transformers for Data Scientists in a rush.
---
This repository contains low-code, easy to understand, pre-built pipelines for fast experimentation on NLP tasks using [huggingface/transformers](https://github.com/huggingface/transformers) pre-trained language models, which are explained and explored in a post series on Medium about the theme.
This was inspired by a LinkedIn post of Thomas Wolf, HuggingFace's CSO in which there was an image of a low-code pipeline for fast experimentation on their Transformers repo. As I could not see anything like it implemented on the internet, I've decided to do it myself.
# Index
As of now, we have:
* [classification](#Classification), with a classification example.
## Classification
On the classification example, we use a [dataset for email spam classification](https://www.kaggle.com/team-ai/spam-text-message-classification) from Kaggle, and use [optuna](https://optuna.org/) for hyperparameter tuning.
You might run it, on the classification directory, with:
```bash
python classification-experiment.py --model-name bert-base-multilingual-cased ---metric f1_score --train-data-path train.csv --test-data-path test.csv --max-sequence-length 25 --label-nbr 2
```
It should yield a f1_score higher than 0.9.
---
###### Made by Pi Esposito