Skip to content
/ LBN Public

TensorFlow implementation of the Lorentz Boost Network (LBN).

License

Notifications You must be signed in to change notification settings

riga/LBN

Repository files navigation

Lorentz Boost Network (LBN) Build Status Package Status

TensorFlow implementation of the Lorentz Boost Network from arXiv:1812.09722 [hep-ex].

LBN architecture

Usage example

import tensorflow as tf
from lbn import LBN

# initialize the LBN, set 10 combinations and pairwise boosting
lbn = LBN(10, boost_mode=LBN.PAIRS)

# create a feature tensor based on input four-vectors
features = lbn(four_vectors)

# use the features as input for a subsequent, application-specific network
...

Or with TensorFlow 2 and Keras:

import tensorflow as tf
from lbn import LBN, LBNLayer

# start a sequential model
model = tf.keras.models.Sequential()

# add the LBN layer
input_shape = (6, 4)
model.add(LBNLayer(input_shape, 10, boost_mode=LBN.PAIRS))

# add a dense layer
model.add(tf.keras.layers.Dense(1024))

# continue builing and training the model
...

For more examples on how to set up the LBN with TensorFlow (eager mode and autograph / tf.function ) and Keras, see this gist.

Installation and dependencies

Via pip:

pip install lbn

NumPy and TensorFlow are the only dependencies. Both TensorFlow v1 and v2 are supported.

Testing

Tests should be run for Python 2 and 3 and for TensorFlow 1 and 2. The following commands assume you are in the root directory of the LBN respository:

python -m unittest test

# or via docker
for tag in 1.15.2 1.15.2-py3 2.1.0 2.2.0; do
    docker run --rm -v `pwd`:/root/lbn -w /root/lbn tensorflow/tensorflow:$tag python -m unittest test
done

Contributing

If you like to contribute, we are happy to receive pull requests. Just make sure to add new test cases and run the tests. Also, please use a coding style that is compatible with our .flake8 config.

Development