Materials and tests to support the course at Düsseldorf University of Applied Sciences (HSD).
It is recommended to create a new environment for this course with many Python libraries that we will use in the Live Coding sessions. You can simply download the environment.yml file in this repository, or clone the repository using:
git clone https://github.com/florian-huber/ai_for_smart_health.git
Then, in the folder with the environment.yml file simply run:
conda update -n base -c defaults conda # optional (to make sure conda is up to date)
conda env create -f environment.yml
This should create a Python 3.10 environment with the packages listed in the yaml-file.
This course will include a lot of machine learning, mostly also deep learning. For this, we will work with the Python libraries scikit-learn (classical machine learning) and Pytorch (deep learning).
Scikit-learn can be accelerated using scikit-learn-intelex (https://intel.github.io/scikit-learn-intelex), although this will not be needed for the live coding sessions we do.
Pytorch can make use of GPUs, which can greatly enhance deep learning model training. For most of the course, however, this will not be necessary.
For this -and the coming- sessions, we will be using the ChestX-ray8 dataset which contains 108,948 frontal-view X-ray images of about 30,000 unique patients.
- Each image in the data set contains multiple text-mined labels identifying 14 different pathological conditions.
- These in turn can be used by physicians to diagnose 8 different diseases.
- In the next sessions we will use this data to develop a single model that will provide binary classification predictions for each of the 14 labeled pathologies.
- In other words it will predict 'positive' or 'negative' for each of the pathologies.
The full dataset is available for free here. No need to download the full dataset for the course! In the course, we will work with a smaller subset!
- We will work with a subset containing about 10% of the original data