AtlasAutoEncoder

The AutoEncoder compress the four-momentum of a sample of simulated particles from 4 to 3 variables.

Here, you can find the codes for implementation of autoencoder-decoder network to compress the data from the ATLAS experiments from 4 to 3 variables, and reconstruct the same.

Data

The ATLAS experiments being conducted at the Large Hadron Collider at CERN produce millions of proton-proton collision events. Here, I analyse the variables from the 4-momentum from the leading jets.

• Training Set
• Test Set

The 4 Variables:

• pt: transverse momentum pT
• eta: pseudorapidity η
• phi: azimuthal angle φ
• eta: energy E

Environment

Python 3.5

Dependencies

• numpy  
• pandas
• matplotlib
• sys
• torch
• fastai == 1.0.61
• scipy
• seaborn
• corner

Code

The Colab Notebook consist of the following modules:

1. Loading the dataset.
   The analysis has been done by preprocessing the data in two ways and comparing the results by auto-ecoding the two sets of preprocessed data. Preprocessing of the data:

• Normalisation of data: Subtract the mean of the training set and divide by standard deviation of the training set.
• Singlular Value Decomposition(SVD): Mathematical technique used for dimensionality reduction. We will add the singular values to original dataframe and then pass it through the deep autoencoder network.
• Utils.py : This file contains basic script used for preprocessing of data like removing delimiters, normalization and svd

2. Histogram to visulaise the data distribution.
3. Train the Model. (NOTE: You can just use the pretrained model by loading it) -Model
4. Reconstruction of the data using the auto encoder-decoder model.
   The plots for both the reconstructed normalised as well as custom standardised data are shown.
5. Reconstruction Loss (Residual).
   Residual = (Predicted Data - Original Data) / Original Data

Usage

1. Install dependencies

2. Downloading the dataset.

Download the datasets and store them in a folder processed_data:

2. Run the Jupyter notebook.

The various plots can be visualised in the Jupyter Notebook

Conclusions

1. Validation Loss : 0.000028

2. Residual : 2.9843 e-05