This repository contains Jupyter notebooks for probabilistic modeling of vibrational spectroscopic datasets. All models have been implemented using the Python probabilistic programming library PyMC3. To use the software, first setup your system by creating a virtual environment containing the required Python packages.

First make sure Python3, pip and graphviz are installed on the system:

>> sudo apt install python3 python3-dev python-pip graphviz

To avoid messing up the Python installation, install virtualenv (https://virtualenv.pypa.io/en/stable):

>> pip install virtualenv

Once installed, create a new virtual enviroment by running the command:

>> virtualenv -p python3 PPSDA

You can activate the virtual environment by running:

>> source PPSDA/bin/activate

To deactivate the virtual environment:

>> deactivate

Activate and cd into the created virtual environment. Install the following Python packages:

>> pip install --upgrade numpy scipy scikit-learn matplotlib pandas jupyter seaborn pymc3 arviz graphviz

To run and experiment with the models, start a Jupyter Notebook server and open the .ipynb files containing the models inside the PPSDA/code/ directories:

>> jupyter notebook

For Windows it is advised to first install the Miniconda environment (https://docs.conda.io/en/latest/miniconda.html). After Miniconda installation, open a shell and create a new environment:

>> conda create --name PPSDA

Enter the new environment by running:

>> conda activate PPSDA

To exit the new environment enter:

>> conda deactivate

Activate the new environment and install the following Python packages:

>> conda install numpy scipy scikit-learn matplotlib pandas jupyter seaborn pymc3

Install the arviz and graphviz libraries:

>> conda install -c conda-forge arviz python-graphviz

To run and experiment with the models, start a Jupyter Notebook server and open the .ipynb files containing the models inside the PPSDA/code/ directories:

>> jupyter notebook

The coffees dataset contains 56 FTIR samples of two coffee species, Arabica (29) and Robusta (27). The spectra were truncated to 800-2000 cm^-1. The dataset was obtained from: https://csr.quadram.ac.uk/example-datasets-for-download/

The juices dataset contains 983 FTIR samples originating from two classes of fresh fruit juices, non-strawberry (632) and strawberry (351). The spectra were truncated to 899-1802 cm^-1. The dataset was obtained from: https://csr.quadram.ac.uk/example-datasets-for-download/

The olive oils dataset contains 120 FTIR samples originating from Spain (50), Italy (34), Greece (20) and Portugal (16), corresponding to four different classes. The spectra were truncated to 799-1897 cm^-1. The dataset was obtained from: https://csr.quadram.ac.uk/example-datasets-for-download/

The wines dataset contains 44 FTIR samples originating from wines produced from the same grape (100% Cabernet Sauvignon), but harvested in different geographical areas, Chile (15), Australia (12), South Africa (11) and Argentina (6). The dataset was obtained from: https://www.models.life.ku.dk/Wine_GCMS_FTIR

The tablets dataset contains a collection of Near-infrared (NIR) and Raman spectra obtained from 4 different types of pharmaceutical tablets with a varying amount of active substance. NIR spectra: 310 samples of type A (70), B (80), C (80) and D (80). Raman spectra: 120 samples of type A (30), B (27), C (33) and D (30). The dataset was obtained from: https://www.models.life.ku.dk/Tablets

The beers dataset contains NIR and Raman spectra of Rochefort 8 (class 1) and Rochefort 10 (class 2) beers. NIR spectra: 44 samples of class 1 (28) and class 2 (16). Raman spectra: 45 samples of class 1 (29) and class 2 (16).