This code provides the ML training & testing methods used in the paper "Machine learning the derivative discontinuity of density-functional theory" (https://arxiv.org/abs/2106.16075)
- Read the requirements. The python version and necessary packages (+versions) which have been used for the ML training & testing are listed here
- Copy all files into an arbitrary directory where you create the new folders TrainingsSets, TestSetsSpecial and Pics.
- Download the Sets.zip from https://zenodo.org/record/5091466#.YOwD-TqxVH4.
It contains the files
- DictTorch_X.0.05-0.20-0.50-0.8-0.95-1.gz -> data used for training, validating and testing
- DensData_H2_11_5.csv -> data used for the prediction of H_2 dissociation curve
- xc_Jump_0.01-0.02-1.csv -> data containg 0.01 fractional densities used to verify the uniform jump of the xc potential
- Extract DictTorch_X.0.05-0.20-0.50-0.8-0.95-1.gz into the TrainingsSets folder, and DensData_H2_11_5.csv and * Vxc_Jump_0.01-0.02-1.csv into the TestSetsSpecial folder
- For training of models via train.py we propose to use a file containing all arguments/hyperparameters - e.g. named specsTrainExample - and to type command
python train.py @specsTrainExample(an example of the file with the same name is already provided) - For testing of models via test.py we propose to use a file containing all arguments as well - e.g. named specsTestExample - and to type command
python test.py @specsTestExample(an example of the file with the same name is already provided) - Type
-hinstead of@specsTrainExampleor@specsTestExamplefor a detailed description of all necessary and optional arguments being available - The files H2Dissociation_TestModel.pdf and VxcJump_TestModel.pdf should be the ouput files obtained by passing the arguments (in specsTrainExample & specsTestExample) for training and testing respectively
- the files m101-m107 in Models.zip containing different hyperparameters correspond to the models presented in the paper