NNimaker 4.0

A module for processing user data to NN input

Neural Network inputs maker

The comprehensive documentation and installation instructions, please visit the Link.

Description

This python package will detect the output of other codes (still Quantum Espresso) and create a json file (called data.json) that can be used as the raw input for Neural Networks.

Prerequisite

The python libraries which are needed to install:

numpy,
pandas,
json2,
collection,
collections.abc,
glob2

Other codes and necessary files:

QE output (mandetory) User have to provide a Quantum ESPRESSO output (PWSCF) as a necessary ingredient for this package! The folder that includes this file, should be specify with argument --folder [directory] in runtime!

Platform or supercomputer (mandetory) Supercomputer should be identified with argument --platform [supercompure name] in runtime too!

There is no other necessary inputs, but optional one!

SYSINFO.JSON (optional!) The user can improve the hardware information from the utilized flatform or supercomputer. This information can be provided as a JSON file with this keywords: "Platform": platform, "CPU": cpu_model, "Node": node_info, "Memory": f"{total_memory_gb:.2f} GB DDR4 RAM" Although in order to avoid error, we recommend to use IDENTIKEEP plugin for generating the JSON file. This JSON file sould be named sysinfo.json (default name for IDENTIKEEP output) and placed in the same directory as of QE output.

How to install

Method 1:

Installation is based on Python Package Index (PyPI) repository.

pip install NNimaker

Method 2:

However, if you prefer to create the wheel of package on your laptop, you just need to download the source code from the origin repository . Then follow the path below to create the wheel on your system and then install it. (Note that you may need use virtual environment to install the necessary packages)

pip install setuptools wheel
python setup.py sdist bdist_wheel
pip install dist/NNimaker-<VERSION>-py3-none-any.whl 

How to USE

It is possible to use terminal commands ( The console point is provided). Since it supports just Quantum Espresso output (till this moment), the command is:

NNimake_QE  --folder [DIRECTORY] --outname [OUTPUT] --platform [SUPERCOMPUTER] --algname [ALGORITHM]

All of the arguments have default values. Default for directory is ./ , Default for output name is data.json, Default for platform is Leonardo-booster and Default for algorithm name is davidson. For example, I put the sysinfo.json and output file of QE in ../work/ directory. The command line would be:

NNimake_QE --folder ../work/ --platform "Leonardo-booster"

You can import the package in your python code directly:

from NNimaker import create_json

create_json("../work", outname="data.json", platform="Leonardo-booster", algoname='davidson')

Then you can find the created JSON file (data.json) in the directory that you specified with flag outname.

You can help us to grow our NN model with your data, if you willing to do so please share your results with us through this google form.

Good luck!