https://github.com/suvoooo/cta-sourceextension2023

https://github.com/suvoooo/cta-sourceextension2023

Last synced: about 1 month ago
JSON representation

• Host: GitHub
• URL: https://github.com/suvoooo/cta-sourceextension2023
• Owner: suvoooo
• Created: 2023-07-03T11:40:34.000Z (over 1 year ago)
• Default Branch: main
• Last Pushed: 2023-08-10T07:19:40.000Z (over 1 year ago)
• Last Synced: 2024-10-28T15:25:45.476Z (3 months ago)
• Language: Jupyter Notebook
• Size: 11.3 MB
• Stars: 0
• Watchers: 0
• Forks: 0
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

        
# CTA-SourceExtension2023

Corresponding conference proceedings [(ICRC 2023; Nagoya, JP)](https://www.icrc2023.org/) is available [here](https://pos.sissa.it/444/599/).

To cite: 

```

@inproceedings{Vodeb:2023eS,

  author = "Vodeb, V  and  Bhattacharyya, S  and  Principe, G  and  Zaharijas, G  and  Ruiz, R  and  Stoppa, F  and  Caron, S  and  Malyshev, D",

  title = "{Investigating the VHE Gamma-ray Sources Using Deep Neural Networks}",

  doi = "10.22323/1.444.0599",

  booktitle = "Proceedings of 38th International Cosmic Ray Conference {\textemdash} PoS(ICRC2023)",

  year = 2023,

  volume = "444",

  pages = "599"

}

```

## Neural Net Architecture (Simplified)

![Neural-Net](https://github.com/suvoooo/CTA-SourceExtension2023/blob/main/plots/plot_neural_net_ICRC2023.png-1.png)

----------------------------------------------

### Activation Maps from Conv4 Layer for Sources with Different Extensions

#### C0: $0.03 < \sigma < 0.1$

![Activation Maps](https://github.com/suvoooo/CTA-SourceExtension2023/blob/main/plots/check_conv_layers_C0_999_final.png)

#### C1: $0.1 < \sigma < 0.3$

![ActivationsMapsC1](https://github.com/suvoooo/CTA-SourceExtension2023/blob/main/plots/check_conv_layers_C1_999_final.png)

--------------------------------------------

### Libraries & Versions: 

1. Python: `3.9.12`

2. Matplotlib: `3.5.1`

3. Numpy: `1.22.3`

4. Scipy: `1.8.1`

5. Sklearn: `1.0.2`

6. TensorFlow: `2.4.1`

7. Gammapy: `1.0.1`

8. ctools: `1.7.4`

----------------------------------------------------

### scripts:  

Python scripts used for the production 

0. _**generate_templates_updated.py:**_ Template generation (fits format) including cosmic-ray background and source contributions using `ctools`.  

1. _**fits_to_npy_CTA_extent_sel.py:**_ Convert the fits files from `ctools` simulation to numpy arrays; For every fits file we have 4 numpy arrays for 4 energy bins

2. _**classification_DataLoader_CTA_Ext.py:**_ Dataloader module used for preprocessing the numpy arrays to make them suitable for our network. Also includes augmentation for the training set.

3. _**neural_nets.py:**_ Module to import the network used for our task.

4. _**train.py:**_ Training script; Train, evaluate (on validation set) and test (on test-set); Plot activation maps from hidden conv layer (for a random test image).  

----------------------------------------------------

### notebooks:

Helping Notebooks to visualize several sections of scripts 

1. _**read_fits_augment_npy.ipynb:**_ visualize the numpy arrays (source images) after processing the fits files; Also a visualizer for the augmentations used.

2. _**Neural-Net-Arch.ipynb:**_ import neural_nets.py and check in detail the network used for analysis.