PANNA-EVAL¶
Purpose of Module¶
PANNA-EVAL module evaluates an all to all connected neural network to predict atomistic quantities, e.g. total energy and forces of a given crystal structure.
PANNA-EVAL can be used with other modules of the PANNA project for neural network validation, but it can also serve to carry the information of the trianed network to other platforms such as molecular dynamics code LAMMPS.
Although PANNA-EVAL does not need the advanced capabilities of the TensorFlow framework, it uses the ‘checkpoint’ information to automatically test the performance of a network from training data.
Features¶
PANNA-EVAL module has two user-end scripts: evaluate.py and extract_weights.py.
The main script of the PANNA-EVAL module, evaluate.py can evaluate all to all connected networks with various sizes for each species.
It can also calculate the derivative of the target function, ie. forces for an energy network.
This module was primarily created to validate TensorFlow networks stored during training in checkpoint format, hence it has the functionality to look for checkpoint numbers in a training directory, and/or run several checkpoint evaluations at once.
The extract_weights.py script allows one to save the network parameters from TensorFlow native checkpoint format to other useful ones, such as
human readable or LAMMPS potential formats. This last one allows neural networks that are trained and validated using PANNA [1] modules to
be exported to LAMMPS as interatomic potentials.
Building and Testing¶
A stable version of the module can be downloaded using the download button on this page
As a python module PANNA-EVAL does not require installation but it relies on numpy library version >= 1.15.0, TensorFlow version >= 1.13.0. Note that with version 2.0.0, TensorFlow libraries went under substantial changes in structure, the 1.1X.X family supports the equally valid previous structure and is still being maintained. PANNA-EVAL requires TensorFlow 1.1X.X family of versions.
In order to set up and test the module, run the following:
$ tar -zxvf panna-master.tar.gz
$ cd panna-master
$ python3 ./panna/test-evaluate.py
Currently this test only assesses the evaluate.py script. Another test for extract_weights.py will be released in the near future.
Usage¶
PANNA-EVAL main script requires a configuration file that specifies the parameter of the calculation such as where to find the network to evaluate, which checkpoints to evaluate etc.. A typical command for using this module is as follows:
$ export PYTHONPATH=/path/to/panna/directory/panna
$ python3 evaluate.py --config val_config.ini
A detailed tutorial about the contents of the configuration file can be found here.
In this comprehensive tutorial, a neural network training scenario is demonstrated from beginning to end. Network training and validation are two key steps of generating a predictive network, hence in the tutorial how to use this module together with PANNA-TRAIN module used in training is also explained. Together, these two modules cover all the steps necessary to train an atomistic neural network, starting from a data which specifies the machine learning task in (input, target output) pair form.
Source Code¶
PANNA-EVAL source is currently hosted on GitLab.
Further Information¶
The PANNA-EVAL module is developed with the contributions of R. Lot, Y. Shaidu, F. Pellegrini, E. Kucukbenli.
References¶
PANNA manuscript:
| [1] | R. Lot, Y. Shaidu, F. Pellegrini, E. Kucukbenli. arxiv:1907.03055. Submitted (2019). |