Skip to content

Files

Latest commit

d1807c8 · Nov 27, 2024

History

History

hgcal_autoencoder

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
Jul 23, 2024
Nov 22, 2024
Jul 23, 2024
May 20, 2024
Nov 27, 2024
Jun 11, 2024
Jun 11, 2024
May 20, 2024
May 20, 2024
May 20, 2024
May 20, 2024

HGCal Autoencoder

This repo contains a small, medium, and large Autoencoder trained on the HGCal dataset. These models were developed using a Bayesian optimization process, and the models were selected along a Pareto front defined by Model error versus Binary Operations (BOPS).

Models provided:

  • Small Pareto
  • Medium Pareto
  • Large Pareto

Dataset

To evaluate the model, download the dataset here.

Load the model

We provide a script to load the model for further evaluation: ./scripts/load_model.sh. No need to download the dataset to load the model.

QKeras versions

To load the Medium Pareto model, run:

./scripts/load_model.sh 0

To load the Small Pareto model, run:

./scripts/load_model.sh 1

To load the Large Pareto model, run:

./scripts/load_model.sh 2

FKeras versions

To load the Medium Pareto model, run:

./scripts/load_model.sh 3

To load the Small Pareto model, run:

./scripts/load_model.sh 4

To load the Large Pareto model, run:

./scripts/load_model.sh 5

Evaluate a model

We provide a script for evaluating the Earth Mover's Distance (EMD) of the model: ./scripts/eval.sh. Lower EMD is better. A perfect autoencoder reconstruction of the data would yield an EMD of 0.

Make sure to point the .scripts/eval.sh's DATASET path to where you have stored the dataset, e.g.,

DATASET=./data/pickled--data_values--phys_values--EoL_dataset.pkl

FKeras versions

To evaluate the Medium Pareto model, run:

./scripts/eval.sh 0

To evaluate the Small Pareto model, run:

./scripts/eval.sh 1

To evaluate the Large Pareto model, run:

./scripts/eval.sh 2