This repository contains the code for the experimentation shown in the paper presented at KDD Workshop on Discovering Drift Phenomena in Evolving Landscapes (DELTA 2024).
Preprint version: link
execute:
conda create -n cpnn python=3.8
conda activate cpnn
pip install -r requirements.txt
The project is composed of the following directories.
It contains the generated data streams. Each file's name has the following structure: <generator>_<configuration>.csv.
Generators:
- sine_rw10: Sine RW (SRW).
- sine_rw10_mode5: Sine RW Mode (SRWM).
- weather: Weather.
Configurations:
- 1conf:
- S1+ S2+ S1- S2- for Sine and Sine RW Mode
- W1+ W2+ W1- W2- for Weather.
- 2conf:
- S1+ S2- S1- S2+ for Sine and Sine RW Mode
- W1+ W2- W1- W2+ for Weather.
- 3conf:
- S2+ S1+ S2- S1- for Sine and Sine RW Mode
- W2+ W1+ W2- W1- for Weather.
- 4conf:
- S2+ S1- S2- S1+ for Sine and Sine RW Mode
- W2+ W1- W2- W1+ for Weather.
It contains the python modules implementing cPNN, cLSTM, cGRU.
It contains the python modules to implement the prequential evaluation used for the experiments.
It contains the python modules implementing the data stream generator.
It runs the prequential evaluation using the specified configurations. Change the variables in the code for different settings (see the code's comments for the details).
Run it with the command python -m evaluation.test.
The execution stores the pickle files containing the results in the folder specified by the variable PATH_PERFORMANCE. For the details about the pickle files, see the documentation in evaluation/prequential_evaluation.py.
https://github.com/AndreaCossu/ContinualLearning-SequentialProcessing