A Machine Learning Platform for the Discovery of Materials

Results and models for the paper A Machine Learning Platform for the Discovery of Materials (Belle, Aksakalli, Russo).

Notes

There are two approaches to use these models.

Online

These models are made available for use via a web interface at https://hadokenmaterials.io/ - pop in the relevant figures and off you go. Easy! Also available is an API which you can call to make predictions. These interactions can be as simple as:

POST

{
	"stoichiometry": "Ca2Cu2Ge4O12"
}

RESPONSE

{
	"bandGap": 1.3985653904114555472784324321,
	"stoichiometry": "Ca2Cu2Ge4O12"
}

NOTE: Any use of https://hadokenmaterials.io/ (web interface or API) is bound by the following citing policy: https://hadokenmaterials.io/Home/Citing - please make a note of this.

Offline

• Install Keras and appropriate other Python bits
• Load the model into Keras using the appropriate H5 and JSON files
• Call the model specifying values for the relevant inputs (note that inputs required for each model differ - relevant features are noted in the corresponding Output.log file)
• Save and load Keras models - https://www.tensorflow.org/guide/keras/save_and_serialize
• python - How to predict from saved model in Keras ? - https://stackoverflow.com/questions/50227925/how-to-predict-from-saved-model-in-keras

Directory Structure

• Data
  • 2020-Feb-16_23-33-16-PAW_PBE_Reduced_HighSymmetry-v5.00.csv - dataset used to generate the models
• Models
  • BandGap_NULL_P
    • BandGap_NULL_T_Results.csv - results of the training process, including original E_g values and fit
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.h5 - Keras model output
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.json - Keras model output
    • History.png - loss vs. epoch plot
    • Output.log - logging information for the entire process
  • BandGap_SpaceGroup-Geometry_P
    • BandGap_SpaceGroup-Geometry_T_Results.csv - results of the training process, including original E_g values and fit
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.h5 - Keras model output
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.json - Keras model output
    • History.png - loss vs. epoch plot
    • Output.log - logging information for the entire process
  • BandGap_SpaceGroup-HighSymmetry-Derived_P
    • BandGap_SpaceGroup-HighSymmetry-Derived_T_Results.csv - results of the training process, including original E_g values and fit
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.h5 - Keras model output
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.json - Keras model output
    • History.png - loss vs. epoch plot
    • Output.log - logging information for the entire process
  • FermiEnergy_Geometry_P
    • FermiEnergy_Geometry_P_Results.csv - results of the training process, including original E_F values and fit
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.h5 - Keras model output
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.json - Keras model output
    • History.png - loss vs. epoch plot
    • Output.log - logging information for the entire process
  • GapType-Classifier_P
    • GapType-Classifier_P_Results.csv - results of the training process, including original Gap Type values and fit
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.h5 - Keras model output
    • EnvT.L100.L50.RELU.RELU.D0.01.E300.B200.Model.json - Keras model output
    • Output.log - logging information for the entire process

Correspondence

All contacts are contained within the paper - please feel free to contact us as you see fit.