install this package as a python module with pip via:

pip install git+https://github.com/erikbuh/erum_data_data.git

The essential function so far is the "load" function to load the training and testing datasets. The datasets features "X" are returned as a list of numpy arrays. The labels are returend directly as a numpy array.

import erum_data_data as edd

# loading training data into RAM (downloads dataset first time)
X_train, y_train  = edd.load('top', dataset='train', cache_dir = './', cache_subdir = 'datasets')

# loading test data into RAM (downloads dataset first time)
X_test, y_test = edd.load('top', dataset='test', cache_dir = './', cache_subdir = 'datasets')

Here a subfolder ./datasets is created. The datasets take up a total disk space of about 2.4 GB. For loading the training datasets a free RAM of at at least 5 GB is necessary (depending on the dataset).

Included datasets at the moment with the tags:

1: 'top', 2: 'spinodal', 3: 'EOSL', 4: 'airshower', 5: 'belle'

An description of the datasets can be printed via the function:

edd.print_description('top')

Some example plots can be found in the notebooks in the example folder.

A simple model implementation can be found in the folder 'simple_model'. To run the notebook one needs to additionally install at least tensorflow version >= 2.0 and scikit >= 0.22.

The original datasets can be found here:

1. Top Tagging at the LHC link, Publication: 1902.09914
2. Spinodal or not? link, Publication: 1906:06562
3. EOSL or EOSQ link, Publication: 1910.11530
4. Cosmic Airshower link
5. SmartBKG dataset (Belle II - generated events passing downstream selection) link