This repository contains data pickels for multilabel classification dataset for easy research. I like to think of this as my mini-contribution. Each directory contains a single datset, with the respective pickles and count.txt which give the dimensions of features and labels in the order.

dataset_name
|____________ dataset_name-train-features.pkl
|____________ dataset_name-test-features.pkl
|____________ dataset_name-train-labels.pkl
|____________ dataset_name-test-labels.pkl

import numpy as np

def get_data(path, noise=False):
    data = np.load(path)
    if noise == True :
        data = data + np.random.normal(0, 0.001, data.shape)
    return data

x_train = get_data("./datset_name/dataset_name-train-features.pkl")

To convert your standard arff files to numpy pickles, which are easier to use and faster to process, use the script I have included along with the datasets.

Download the dataset and put it in this folder. Create a count.txt in the directory and put the dimensions in the following format :

features_dim
labels_dim

pip3 install liac-arff 
python3 to_numpy --dataset dataset_name