/perovskite-oxide-stability-prediction

code package with elemental property dictionary that trains a model based on training dataset and gives prediction on new perovskite compounds

Primary LanguagePython

This package can give predictions on stability of new perovskite compoounds using machine learning approach.

How to use:

  1. New compounds info should be put in the newCompound.xlsx file, with the specified format as described in the excel file
  2. The training data should be put in the perovskite_DFT_EaH_FormE.xlsx. This file already contains over 1900 compounds. Users can add more compounds to this training set.
  3. Run python script energy_prediction.py using command: python energy_prediction.py. Requirements:

  • python version >= 3.5

  • sklearn, pandas, numpy.

  1. During the running of the python script, temporary files *.csv will be generated, which contain the generated features of the compounds in the training set and the new compounds to be predicted. The temporary files are generated by feature_vector.py, called by the energy_prediction.py.
  2. Results are put in the prediction_result.xlsx file. In the stability column, 1 stands for stable, 0 stands for unstable.

Information:

The energy_prediction embeds the selected best classification model (extra trees) and the best regression model (kernel ridge regression).

The elemental property database is provided as continousproperty.xlsx, discrictproperty.xlsx and shannon_perovskite.xlsx. Feature inportance list is stored by *.txt files.