AcRoPLS :classical_building:

AcRoPLS (Acurate Robust PLS) provides a methodology to create accurate and robust PLS models based on solving a data-driven optimization problem that couples data pre-processing and model regression to a single optimization step. The accuracy objective is evaluated based on the performance of the generated model on predicting the model output on a test set. For the robustness objective, a novel metric based on the method of moments applied for different realizations of a known variability source evaluated again on a test set. <!-- For more information on the method, please refer to: [add paper link](). -->

Requirements

The optimization is performed using ENTMOOT and the PLS model is created using pyphi. Both these packages are required for the code to be excecuted. All code was run on python 3.8.5 with following package versions:

Packages

|Package| Version| |-------|--------| |cycler | 0.10.0| |matplotlib | 3.4.1| |numpy | 1.20.2| |pandas | 1.2.4| |scikit-learn | 0.24.2| |scipy | 1.6.3| |openpyxl|3.0.10|

ENTMOOT Installation

This requires version 0.2.2 of ENTMOOT. See list of requirements from https://github.com/cog-imperial, then install 0.2.2 of ENTMOOT using

pip install git+https://github.com/cog-imperial/entmoot.git@8e6e0fbfaba7823269065ee3c88d56d354167bb9

Running Examples

Navigate to the examples file in your terminal and run the desired example, e.g.

python3 InitialExample.py

<!--Note that the third example in [add paper link]() is not provided in this repository due to confidentiallity of the industrial dataset used for that particular case study.-->

Contributors

| Contributor | Acknowledgements | | ---------------- | ------------------------- | | Chryssa Kappatou | This research is funded by an Engineering and Physical Sciences Research Council / Eli Lilly Prosperity Partnership (EPSRC EP/T005556/1) and by Eli Lilly & Company|