Section MACHINE_LEARNING

• Machine learning section

• Section path: CP2K_INPUT / FORCE_EVAL / DFT / LS_SCF / PAO / MACHINE_LEARNING

• This section cannot be repeated.

• Subsections

  • TRAINING_SET

• Keywords

  • DESCRIPTOR
  • GP_NOISE_VAR
  • GP_SCALE
  • METHOD
  • PRIOR
  • TOLERANCE

• Keyword descriptions

  DESCRIPTOR
  	DESCRIPTOR {Keyword}
  	Descriptor used as input for machine learning.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: POTENTIAL
  	List of valid keywords: OVERLAP Eigenvalues of local overlap matrix POTENTIAL Eigenvalues of local potential matrix R12 Distance between two atoms (just for testing)
  	This keyword cites the following reference: [Zhu2016]

  GP_NOISE_VAR
  	GP_NOISE_VAR {Real}
  	Variance of noise used for Gaussian Process machine learning.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-001

  GP_SCALE
  	GP_SCALE {Real}
  	Length scale used for Gaussian Process machine learning.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 5.00000000E-002

  METHOD
  	METHOD {Keyword}
  	Machine learning scheme used to predict PAO basis sets.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: GAUSSIAN_PROCESS
  	List of valid keywords: GAUSSIAN_PROCESS Gaussian Process LAZY Just rely on prior NEURAL_NETWORK Neural Network

  PRIOR
  	PRIOR {Keyword}
  	Prior used for predictions.
  	This keyword cannot be repeated and it expects precisely one keyword.
  	Default value: ZERO
  	List of valid keywords: MEAN Use average of training-set ZERO Simply use zero

  TOLERANCE
  	TOLERANCE {Real}
  	Maximum variance tolerated when making predictions.
  	This keyword cannot be repeated and it expects precisely one real.
  	Default value: 1.00000000E-002