MLHEP2020: Calorimeter regression

Please log in.

    • MLHEP 2020 Calorimeter regression

      The MC dataset used in this competition was generated with the code kindly provided by the authors of EPJ Web of Conferences 214, 02034 (2019), Chekalina et al. 

      This challenge is aimed to recover the particle initial position and momentum in LHCb calorimeter. 

      The calorimeter system is designed to stop particles as they pass through the detector, measuring the amount of energy lost as each one grinds to a halt. Two types of calorimeter are used at LHCb.
       
      The electromagnetic calorimeter is responsible for measuring the energy of lighter particles, such as electrons and photons, while the experiment's hadron calorimeter samples the energy of protons, neutrons and other particles containing quarks. 
       

       LHCb detector

                             LHCb detector

      There are two ways of submitting:

      • result submission (just upload a zip with the prediction npz file)
      • code submission (submit code that runs your model and outputs a prediction)

      See the starting kit for an example.

      In order to make code submission please follow the instructions below:

      1. Register at Coopetition, with the same email you have on github or authenticate via github website;
      2. Fork baseline repository;
      3. Follow these instructions to enable automatic solution submission. The {Competition_id} equals to 98, you can always check it in URL;
      4. Update the CompetitionStartingKit.ipynb file at your local copy of the repositoty and push it back to github. The submission should be processed by coopetition automagically;
      5. Have a look at the coopetition leaderboard. There you can find the solutions of others that may serve as a good source of inspiration.

      Good luck!

       
    • MLHEP 2020 Calorimeter regression: Evaluation

      The submission will be evaluated using the following code:

      def scoring_function(solution_file, predict_file):
          score = 0.
          
          solution = np.load(solution_file, allow_pickle=True)
          predict = np.load(predict_file, allow_pickle=True)
          ParticleMomentum_sol = solution['ParticleMomentum'][:, :2]
          ParticlePoint_sol = solution['ParticlePoint'][:, :2]
          
          ParticleMomentum_pred = predict['ParticleMomentum'][:, :2]
          ParticlePoint_pred = predict['ParticlePoint'][:, :2]
          
          score += np.sum(np.square(ParticleMomentum_sol - ParticleMomentum_pred).mean(axis=0) / np.square(ParticleMomentum_sol).mean(axis=0))
          score += np.sum(np.square(ParticlePoint_sol - ParticlePoint_pred).mean(axis=0) / np.square(ParticlePoint_sol).mean(axis=0))
          return score
      

       

    • MLHEP 2020 Calorimeter regression: Rules

      Zipped results must be submitted before the 2020-08-24 23:59:00+00:00.

      Zipped file should contain a single npz-file named data_test_prediction.npzThis npz-file should contain two numpy arrays in it: ParticlePoint and ParticleMomentum. You may submit 10 submissions every day and 200 in total.

      Also, please, if you are submitting not code solution, but raw predictions, please, add a link to where we can find your code in a comment to submission.

    • MLHEP 2020 Calorimeter regression: Data

      Dataset is in npz-format and could be opened with `np.load`. You are provided with two files: data_train.npz and data_test.npz. data_train.npz consists of four numpy-arrays: EnergyDeposit, ParticlePoint, ParticleMomentum, ParticlePDG. data_test.npz contains only one numpy-array: EnergyDeposit.  For an example of how to work with it take a look on the starter-kit. 

    • Development

      starting_kit public_data
  • Make your submission using github

    ID Status Inputs