Machine Learning and Pattern Recognition Project

Authors: Michele Cerra, Giuseppe Stracquadanio

Project Structure for the Delivery

• main.py script: testing performances of all the models on the val. set (K-fold CV)
• evalMain.py script: testing performances of all the models on the eval. set
• calibrationFusionMain.py: performing score calibration and fusion for the best models on the val. set
• finalMain.py: testing the optimal model(s) on the evaluation set

Pipeline Abstraction

For better managing all the models and all the steps, we employed the abstraction of a pipeline. Each model (classifiers.py) or pre-processing step (preProc.py) is a pipeline stage, which can be specified when creating a new pipeline. Then, the pipeline can be received as input by a CrossValidator object, in order to execute all the steps while using the K-fold protocol.