PardFROCC (and other versions) adheres to sklearn estimator interface. PardFROCC and Sparse DFROCC expects sparse matrices in CSC format. Other methods expect numpy arrays. The arrays can be float or int types.

import pardfrocc
from sklearn.metrics import roc_auc_score

x_train, x_test, y_test = load_data() # any function to load data in appropriate format
clf = pardfrocc.ParDFROCC()
clf.fit(x_train)

scores = clf.decision_function(x_test)
roc = roc_auc_score(y_test, scores)

predictions = clf.predict(x_test)

Any numeric data file can be converted to required format as follows:

import numpy as np
import scipy.sparse as sp

x = np.array(x, dtype=np.float32) # for dense data
x = sp.csc_matric(x, dtype=np.float32) # for sparse data

For more information of data loading, see [Sparse matrix docs] and [Numpy array docs].

We provide a script for generating HiMoon and MMGauss dataset.

import data_gen
x_train, y_train, x_val, y_val, x_test, y_test = data_gen.himoon(n_samples=1000, n_dims=1000) #or data_gen.mmgauss()

Parameters:

n_samples - Total number of generated samples (Default: 1000)

n_dims - Number of dimensions of generated data (Default: 1000)

sparsity - Sparsity of the generated data (Default: 0.01)

python experiment.py --dataset <DATASET> --epsilon <EPSILON> --dimension <DIMENSION> --n_samples <N_SAMPLES> --n_dim <N_DIM> --method <METHOD> --repetitions <REPETITIONS> --outfile <OUTFILE>

where

dataset - one amongst mmgauss and himoon

epsilon - Seperation parameter. Typically of the order 0.1-0.0001 (Default: 0.01)

dimension - Number of FROCC dimensions (Default: 1000)

kernel

n_samples - Number of samples to generate when using generated data (Default: 1000)

n_dim - Number of dimensions in the data to be generated when using generated data (Default: 1000)

method - One amongst frocc, dfrocc, sparse_dfrocc, pardfrocc (Default: pardfrocc)

repetations - Number of repetitions to run the experiment for (Default: 1)

outfile - File path to write results to (Default: results/out.csv)

Example:

python experiment.py --dataset himoon --epsilon 0.01 --dimension 100 --n_samples 100000 --n_dim 100000 --method pardfrocc --repetitions 1 --outfile himoon_results.csv

Details of hyper-parameters used to reproduce the results is provided in hyper.pdf

This is released as a research prototype. It is not meant to be a production quality implementation. It has been made open source to enable easy reproducibility of research results.