Create a conda environment using usp_environment.yml file and activate it.
conda env create -f usp_environment.yml
conda activate usp
You will need to set the python path in the following way:
PYTHONPATH=<FULL_PATH_TO_REPO>/undersampling-project/scripts/
where FULL_PATH_TO_REPO is the full path to the repository on your machine.
To run an experiment for MLP, NB, and LR classifiers, run the experiment.py script. Here is its usage:
usage: experiment.py [-h] -n NOISE_DATASET -s SIGNAL_DATASET [-i SNR_INT [SNR_INT ...]] [-f N_FOLDS] [-nn NUM_NOISE] [-v]
options:
-h, --help show this help message and exit
-n NOISE_DATASET, --noise-dataset NOISE_DATASET
noise dataset name (FBIRN, COBRE, VAR)
-s SIGNAL_DATASET, --signal-dataset SIGNAL_DATASET
signal dataset name (OULU, HCP)
-i SNR_INT [SNR_INT ...], --snr-int SNR_INT [SNR_INT ...]
upper, lower, step of SNR interval
-f N_FOLDS, --n-folds N_FOLDS
number of folds for cross-validation
-nn NUM_NOISE, --num_noise NUM_NOISE
number of noise iterations
-v, --verbose turn on debug logging
The SVM experiments need to use a different environment. Create the thundersvm environment from the thundersvm_environment.yml file then activate it.
conda env create -f thundersvm_environment.yml
conda activate thundersvm
You will need to set the python path in the following way:
PYTHONPATH=<FULL_PATH_TO_REPO>/undersampling-project/scripts/
where FULL_PATH_TO_REPO is the full path to the repository on your machine.
To run an SVM classification experiment, you will need the model weights. You can generate them on your own using a hyper parameter search, or they are available by request.
To run a hyperparameter search, run the hps_thundersvm.py script. Here is its usage:
usage: hps_thundersvm.py [-h] -n NOISE_DATASET -s SIGNAL_DATASET -k {linear,rbf} [-i SNR_INT [SNR_INT ...]] [-f N_FOLDS] [-v VERBOSE]
options:
-h, --help show this help message and exit
-n NOISE_DATASET, --noise-dataset NOISE_DATASET
noise dataset name
-s SIGNAL_DATASET, --signal-dataset SIGNAL_DATASET
signal dataset name
-k {linear,rbf}, --kernel-type {linear,rbf}
type of SVM kernel
-i SNR_INT [SNR_INT ...], --snr-int SNR_INT [SNR_INT ...]
upper, lower, step of SNR interval
-f N_FOLDS, --n-folds N_FOLDS
number of folds for cross-validation
-v VERBOSE, --verbose VERBOSE
turn on debug logging
To run an SVM experiment, use the nopoly_thundersvm.py script. Here is its usage.
usage: nopoly_thundersvm.py [-h] -n NOISE_DATASET -s SIGNAL_DATASET -k {linear,rbf} [-i SNR_INT [SNR_INT ...]] [-f N_FOLDS] [-nn NUM_NOISE] [-v VERBOSE]
options:
-h, --help show this help message and exit
-n NOISE_DATASET, --noise-dataset NOISE_DATASET
noise dataset name
-s SIGNAL_DATASET, --signal-dataset SIGNAL_DATASET
signal dataset name
-k {linear,rbf}, --kernel-type {linear,rbf}
type of SVM kernel
-i SNR_INT [SNR_INT ...], --snr-int SNR_INT [SNR_INT ...]
upper, lower, step of SNR interval
-f N_FOLDS, --n-folds N_FOLDS
number of folds for cross-validation
-nn NUM_NOISE, --num_noise NUM_NOISE
number of noise iterations
-v VERBOSE, --verbose VERBOSE
turn on debug logging
For results, see the plot.ipynb notebook. You can modify this script to plot results of your classification experiments.