PANNA-TFR module demonstrates how to efficiently pack the Behler-Parinello and modified Behler-Parinello descriptor vectors (See References , , ) written in binary format, into TensorFlow data format for efficient reading during training.
These descriptors can then be used within TensorFlow efficiently, reducing the overhead during batch creation. PANNA-TFR is built on TensorFlow.
PANNA-TFR supports descriptors that change size across records, i.e. data points with different number of atoms are stored efficiently without padding.
A stable version of the module can be downloaded using the download button on this page
As a python module PANNA-TFR does not require installation but it relies on numpy library version => 1.15.0 and tensorflow version => 1.13.0.
In order to set up and test the module, run the following:
$ tar -zxvf panna-master.tar.gz $ cd panna-master $ python3 ./panna/test-tfr-packer.py
PANNA-TFR main script requires a configuration file that specifies the parameter of the calculation such as location of descriptor files or how many descriptors to be packed in a single record file. A typical command for using this module is as follows:
$ export PYTHONPATH=/path/to/panna/directory/panna $ python3 tfr_packer.py --config tfr_configuration.ini
A detailed tutorial about the contents of the configuration file can be found here.
In this comprehensive tutorial, how use this module with other modules such as PANNA-GVECT and PANNA-TOOLS is also demonstrated. Together, these modules cover all the steps necessary while going from raw data to descriptors that can be used in machine learning workflow.
The PANNA-TFR module is developed with the contributions of R. Lot, Y. Shaidu, F. Pellegrini, E. Kucukbenli
|||R. Lot, Y. Shaidu, F. Pellegrini, E. Kucukbenli. arxiv:1907.03055. Submitted (2019).|
|||J. Behler and M. Parrinello, Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces, Phys. Rev. Lett. 98, 146401 (2007)|
|||Justin S. Smith, Olexandr Isayev, Adrian E. Roitberg. ANI-1: An extensible neural network potential with DFT accuracy at force field computational cost. Chemical Science,(2017), DOI: 10.1039/C6SC05720A|