The main goal of this project is to apply Quantum Kernel Machine learning tecniques to a data classification problem in High Energy Physics; description of the problem and the dataset can be found here
The project relies on the following python packages:
-
qiskit: IBM's quantum computing framework -
matplotlib,seaoborn: for plotting graphs -
numpyandsklearn: for the numerical heavy lifting, the machine learning algorithms and metrics used -
a lot of patience: training and testing this kind of algorithm is very very time consuming, the system on which was tested is anIntel Core i7 6700HQ - 8GB RAM, running the entire test for 1 run takes more or less 1 hour and a half
The repository contains 3 main components:
-
QSVM.ipynbExperimental notebook used to test code -
whole.pyPython File for running the entire project -
utils.pyUtility file with functions for plotting and retriving feature maps
Pareser is not yet implemented, parameters are hard coded in the first lines of the python file.
IMPORTANT: change data_path variable. The file expects an numpy array which has the first k columns as features and the last one as label.
The imported array is then shuffled and divided in 80% for training and 20% for test.
Parameters:
-
n_c: number of features to consider for training and tesing models -
encoders: encoders used to reduce data dimensionality -
C_SVM_SAMPLES: samples used to train the classical model -
Q_SVM_SAMPLES: samples used to train the quantum model -
PREDICTIONS: samples used to test models -
RUNS: number of runs to perform -
PROBA: control the fitting of svm to later retrieve probability score and AUC
Once set all parameters run with:
python3 whole.py
The script outputs confusion matrices for all models tested and a final dump of the auc values and training time.
On the terminal it writes the time required to train and test the models and auc scores of the model tested if PROBA parameter is set to True.
Keynote presentation can be found in keynote_aqm.pdf, unfortunately in Italian.