ZooplanktonClassifierPrototype

A simple prototype to feed in .PID files to a pre-trained classifier. In its current incarnation, a pre-trained XGBoost model is used.

Usage

1. Open RStudio by double-clicking the .Rproj file (or any other R IDE that supports root dir determination via .Rproj).
2. Source/run the Main.R and follow the on-screen prompts:
   1. Point to an .RDS file containing the pre-trained classifier (an XGBoost model is supplied in the Classifier folder).
   2. Point to a folder containing the .pid files (example files are provided in teh SamplePID folder).
3. At the script's end, the results will be save on the same folder as the .pid files but in .csv format, keeping the same name. This allows for easy separation of results and input. For each xyz.pid file the following are generated:
   1. A xyz_predictions.csv containing the !Item and Label identifiers along with the generated predictions.
   2. A xyz_predictionsSummary.csv containing the predicted classes along with the number of predictions corresponding to each predicted class, i.e. how many observations are (estimated to be) of each class.

Overview

The key functionality is the automatic detection of the [Data] flag inside .pid files via regular expressions. The user does not have to provide the line number of where that flag is located, especially useful given that it is not the same in all .pid files.

Furthermore, the script is able to distinguish between .pid and other files, targeting the .pid exclusively.

Otherwise, the script simply formats the data as expected by the classifier (removal of features unrelated to the classification, adding derived features etc.). The original .pid files are left unchanged on disk, all changes are taking place in memory.

The classifier used in this incarnation has been created by George Kampolis with a process fully documented as part of his MSc. project in collaboration with Marine Scotland:

Kampolis Georgios, 2019. Automated Zooplankton Classification, Aberdeen: Robert Gordon University.