This file contain all informations about the project:

Description of the github structure:

• data : folder containing csv of jets events from 2011 to 2016, csv of non jet events from HEK database and the whole downloaded dataset of image sequencies.

• data collection : folder containing a jupyter notebook that shows how the dataset was created and a python file that contain helper functions for this task.

• model training : folder containing a jupyter notebook that shows how the model has been implemented and trained. Also a python file with helper functions for this task.

• model evaluation : folder that contains:

  • model_analysis.ipynb : jupyter notebook that shows the result of the model
  • helper_analysis.py : python file that contain helper functions for this task
  • results_cv_final.json : json file that contain the results of the cross validation (18 models)
  • 2 Best resulting models : Trained_RCNN.pth and Trained_RCNN_2.pth
  • figures : folder that contains every missclassified events as gif, the plots displayed in the notebook

• try_model_with_your_data.ipynb : jupyter notebook for the ESA that will try new datas with our model to benefit from automatic classification and so, detection.

• animation.gif : exemple of the event 29 containing a jet and used in our paper.

Libraries used in the project:

The following is a summary of the libraries and their specific modules used in this project, organized by their functional category:

• Core Libraries

  • NumPy
  • Pandas

• Solar Physics

  • Sunpy
  • Astropy

• Operating System Interaction

  • OS
  • System

• Machine Learning Frameworks

  • PyTorch
  • Torchvision

• Cross-Validation and Metrics

  • Scikit-Learn
  • SciPy

• Visualization and Display

  • Matplotlib
  • IPython Display Utilities
  • Seaborn

• Miscellaneous

  • Random

Run the code

Take care to not run cells in red (These are explicitly noted in the notebooks) because the computationnal requirement is huge.

Steps :

• Make sure to have all used libraries installed.
• Start with data collection
• Then with model training
• Finally with the model evaluation.

Timeline of the Project

• 6-12 November: familiarization with the project and exploration of solar physics Python libraries
• 13-30 November: development of the algorithm to download the data and save it in the correct format
• 1-17 December: design of machine learning architecture and initiation of report writing.
• 18-21 December: completion of results analysis and finalization of the report.