hasnainmamdani / disease_classification Goto Github PK

This is the deliverable of Task 2 that implements a fully functional ML package around a multi-task prediction task.

The goal is to build a model that can predict probability of certain diseases from a genome sequence. The dataset consists of three different types of genome sequences corresponding to three different species – human, duck, and snake. For each species, there is a different set of diseases that they could have – i.e. for each species, a set of probabilities for all possible diseases an organism of that species could have is predicted.

A multi-task prediction model is achieved by using a neural network model with multiple predictor heads.

1. Create a source environment, and install the required dependencies as specified in requirements.txt. This project is tested on python 3.8.
2. Specify the path of dataset in arguments (see run.py)
3. Execute python -m src.run at the root. Model will start training for one epoch and the state will be stored automatically by the PyTorchLightning framework in the folder lightning_logs under root after the training epoch ends.
4. Test cases are under tests/ and can be run with Pytest at the root.