GANs for PDF replicas
Project description
GANPDFs
Enhance the statistics of a prior PDF set by generating fake PDF replicas using Generative Adversarial Neural Networks (GANs). Documentation is available at https://n3pdf.github.io/ganpdfs/.
How to install
To install the ganpdfs
package, just type
python setup.py install or python setup.py develop (if you are a developper)
The package can be installed via the Python Package Index (PyPI) by running:
pip install ganpdfs --upgrade
How to run
The code requires as an input a runcard.yml
file in which the name of the PDF set and the
characteristics of the Neural Network Models are defined. Examples of runcards can be found
in the runcard
folder.
ganpdfs runcard/reference.yml [-t TOT_REPLICAS_SIZE]
In case one does not want to train the GANs and directly resort to a pre-trained one, a pre-trained
model
can be used out of the box by setting the entry use_saved_model
to True
in the runcard.
In order to evolve the generated output grids, just run:
evolven3fit <PRIOR_PDF_NAME>_enhanced <TOT_REPLICAS_SIZE>
Then, to link the generated PDF set to the LHAPDF data directory, use the postgans
script by
running:
postgans --pdf <PRIOR_PDF_NAME> --nenhanced <TOT_REPLICAS_SIZE>
Hyper-parameter opitmization
For more details on how to define specific parameters when running the code and on how to perform a hyper-parameter scan, please head to the section how to of the documentation.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for ganpdfs-1.1.0.dev0-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | b732cec67b8711e42b0096e83723c65f639ae78bb427a404005e0e5559569c7d |
|
MD5 | 3c4b5828b9b560e1d778e1f85b66b24f |
|
BLAKE2b-256 | 43fcf1b0525e5d4b69f86b9405e727fd3bed312a332f7165351a5bc1ac27ffeb |