A [[Python]] package for [[Machine Learning (ML) and AI]] built on top of [[pytorch]].
## Training a Model
![[2024-11-26_fastai.png]]
Each of the five parameters above are crucial. The details are here: www.docs.fast.ai. This creates `dataloaders`, which are iterators that contain trained and predicted data.
The `dataloaders` can be used to train various models. There are built-in models as part of `fastai`, or you can integrate with models here: www.timm.fast.ai/model_architectures. The models are pre-trained, then adjusted based on the data you pass in. This is called fine-tuning.
- `item_tfms` - transformations applied to the items. For images:
- Squish
- Crop
- Pad
- RandomResizedCrop
Use your trained model to identify the data that requires cleaning by looking at images with the highest loss.
## Deploying a Model
A model can be deployed using [[Hugging Face]] and [[Gradio]] or [[Streamlit]].
# References
- www.docs.fast.ai
- Additional available models: www.timm.fast.ai/model_architectures
- [GitHub - fastai/fastbook: The fastai book, published as Jupyter Notebooks](https://github.com/fastai/fastbook)
- Kaggle Notebooks:
- [Is it a bird? Creating a model from your own data | Kaggle](https://www.kaggle.com/code/jhoward/is-it-a-bird-creating-a-model-from-your-own-data)