This repository regroups works on the "Practical dip into deep learning - a PyTorch short crash-course" course of the SIB.
Its goal is to propose a practical introduction to pytorch and deep-learning models by reproducing or getting inspiration from a number published deep-learning models in the field of biology.
This course eschew the theory (which is covered in another SIB course) and focuses instead on the details of the implementation.
The course is targeted to life scientists who are already familiar and fluent with the Python programming language and who have a solid knowledge in machine learning.
In order to follow the course you need to have installed python and jupyter notebooks, as well as a number of prerequisite libraries.
See the intructions on installing prerequisite libraries of more details.
The course is organized in several, numbered, jupyter notebooks, each corresponding to a model which interleaves code demo, and exercises.
The course does not require any particular expertise with jupyter notebooks to be followed, but if it is the first time you encounter them we recommend this gentle introduction.
- 01 my_first_pytorch_neural_network: a very basic Neural Network on synthetic data. The "Hello World!" of pytorch.
- 02 protein_subcellular_localisation_classifier: classical Deep Learning architecture for classification
- 03 single_cell_autoencoder: An autoencoder with a custom loss function
- 04 chest_Xray_CNN: Convolutional Neural Network for image classification
- 05 chest_Xray_Transfer_Learning: Leveraging pre-existing models
Solutions to each practical can be found in the solutions/ folder and should be loadable directly in the jupyter notebook themselves.
Note also the pytorchtools.py file which contain some early stopping utilities.
- data : contains the datasets
- images : images generated or used in the notebooks
- drafts : some drafts notebooks, with many failed attempts...
- solutions: exercise solutions