How to get training data

[mk123qwe]

Dear eric，
How to get training data?I want to find benchmark dataset to confirm the algorithm's effectiveness.
I'm not a physicist, but I'm focused on deep learning.Could you give me some advice?
Thanks！

[jmduarte]

Training data is originally from here in a slightly different format: http://opendata.cern.ch/record/12102

We converted this into our own format for training/ testing.

These files are available on CERNbox.

Training: https://cernbox.cern.ch/index.php/s/HFDzKmSMG1N2hlM
Testing: https://cernbox.cern.ch/index.php/s/pEEleqZN1UlfdZ9

[mk123qwe]

Training data is originally from here in a slightly different format: http://opendata.cern.ch/record/12102

We converted this into our own format for training/ testing.

These files are available on CERNbox.

Training: https://cernbox.cern.ch/index.php/s/HFDzKmSMG1N2hlM
Testing: https://cernbox.cern.ch/index.php/s/pEEleqZN1UlfdZ9
I find that the structure of the h5 is as follows：
spectator_subgroup/spectators
target_subgroup/target
training_subgroup/training_0
training_subgroup/training_1
training_subgroup/training_2
training_subgroup/training_3
weight_subgroup/weights
but，I can‘t recognise the meaning of the each dataset.
training_2 may be 60 charged particle features. The interaction network uses all of the features?
and in test dataset,there are one more mensandnorms.h5?
Could you give me some advise?
THKANS!

[eric-moreno]

spectator_subgroup/spectators - Features containing physics information for the events but not used in the training either because we were trying to decorrelate from them or because they weren't useful. Specific features described in IN_eval.py.

target_subground/target - desired binary classification for training/testing datasets.

training_subground/training_0 - Higher level features from table 4 (https://arxiv.org/pdf/1909.12285.pdf). Not used in IN but still included in the dataset. Specific features also described in IN_eval.py.

training_subgroup/training_1 - Particle flow candidate features (charged and neutral particles). Not used in IN but still included in the dataset. Specific features described in IN_eval.py.

training_subgroup/training_2 - Charged particle tracks from table 2 (https://arxiv.org/pdf/1909.12285.pdf). All features used as input to IN. Specific features also described in IN_eval.py.

training_subgroup/training_3 - Secondary vertex features from table 3 (https://arxiv.org/pdf/1909.12285.pdf). All features used as input to IN. Specific features also described in IN_eval.py.

weights_subgroup/weights - Weights for specific events. Not used in IN but still included in the dataset.

meansandnorms.h5 - Also not used in IN but included in the dataset. Only needed for standardizing inputs if necessary.

[mk123qwe]

spectator_subgroup/spectators - Features containing physics information for the events but not used in the training either because we were trying to decorrelate from them or because they weren't useful. Specific features described in IN_eval.py.

target_subground/target - desired binary classification for training/testing datasets.

training_subground/training_0 - Higher level features from table 4 (https://arxiv.org/pdf/1909.12285.pdf). Not used in IN but still included in the dataset. Specific features also described in IN_eval.py.

training_subgroup/training_1 - Particle flow candidate features (charged and neutral particles). Not used in IN but still included in the dataset. Specific features described in IN_eval.py.

training_subgroup/training_2 - Charged particle tracks from table 2 (https://arxiv.org/pdf/1909.12285.pdf). All features used as input to IN. Specific features also described in IN_eval.py.

training_subgroup/training_3 - Secondary vertex features from table 3 (https://arxiv.org/pdf/1909.12285.pdf). All features used as input to IN. Specific features also described in IN_eval.py.

weights_subgroup/weights - Weights for specific events. Not used in IN but still included in the dataset.

meansandnorms.h5 - Also not used in IN but included in the dataset. Only needed for standardizing inputs if necessary.
Thank you for your detailed explain.