Pinned Repositories
CompressedSensing4MaterialsScience
A machine-learning tutorial on how to identify materials descriptors. The focus is laid on using the sure independence screening and sparsifying operator (SISSO) for regression.)
Chemical-Transferable-Potentials
A python module for building Chemical-tranferabel machine-learning potentials (CTPs). CTPs are models for atomistic simulations. The compound dependent parameters of 2b+3b potentials are described by some neural network. The models approximate the Born-Oppenheimer potential-energy suraface using density-functional-theory calculations as training data. The 2b+3b descriptor and potential module can be accesed also independently of the CTP approach. Publication is in preparation.
ML-and-LA-from-scratch
Some machine learning (ML) methods plus related numerical linear algebra (LA) and also quadratic programming algorithms to solve the ML optimization problems, i.e. no implementation from external libraries is used except for numpy arrays and basic numpy operations on arrays, such as algebraic operations, matrix multiplication, etc. The algorithms are described in theoretical-background.pdf. Note that, neither the implementations are optimized nor are the chosen (especially LA) algorithms to solve the ML problems optimal. The considered LA algorithms are: QR-decomposition based on Gram-Schmidt proces and Housholder reflections, QR algorithm to determine eigenvalues (and vectors for symmetric matrices), singular-value decompostion, Cholesky decomposition, and forward and backward substitution. Quadratic programming is performed via the primal-dual interior-point method. The included machine-learning methods are: linear (least-squares and ridge) regression, non-negative least squares regression, orthogonal matching pursuit, kernel ridge regression, support vector machines, logistic regression, and principal component analysis.
OrthoBasis
A module for orthonormalizing an arbitrary set of one-dimensional functions within an arbitrary interval using the Gram-Schmidt process and numerical integration.
OperableVersion
Class to make variables of type version (e.g. '1.2.1') comparable with each other (e.g. using '>=') as well as allow arithmetic operations (e.g. '+') on them.
TrendAnalyzer
Algorithms to cluster time series and quantify if some cluster has strong trend.
deployment-test
ahmetcik's Repositories
ahmetcik/TrendAnalyzer
Algorithms to cluster time series and quantify if some cluster has strong trend.
ahmetcik/OperableVersion
Class to make variables of type version (e.g. '1.2.1') comparable with each other (e.g. using '>=') as well as allow arithmetic operations (e.g. '+') on them.
ahmetcik/ML-and-LA-from-scratch
Some machine learning (ML) methods plus related numerical linear algebra (LA) and also quadratic programming algorithms to solve the ML optimization problems, i.e. no implementation from external libraries is used except for numpy arrays and basic numpy operations on arrays, such as algebraic operations, matrix multiplication, etc. The algorithms are described in theoretical-background.pdf. Note that, neither the implementations are optimized nor are the chosen (especially LA) algorithms to solve the ML problems optimal. The considered LA algorithms are: QR-decomposition based on Gram-Schmidt proces and Housholder reflections, QR algorithm to determine eigenvalues (and vectors for symmetric matrices), singular-value decompostion, Cholesky decomposition, and forward and backward substitution. Quadratic programming is performed via the primal-dual interior-point method. The included machine-learning methods are: linear (least-squares and ridge) regression, non-negative least squares regression, orthogonal matching pursuit, kernel ridge regression, support vector machines, logistic regression, and principal component analysis.
ahmetcik/Chemical-Transferable-Potentials
A python module for building Chemical-tranferabel machine-learning potentials (CTPs). CTPs are models for atomistic simulations. The compound dependent parameters of 2b+3b potentials are described by some neural network. The models approximate the Born-Oppenheimer potential-energy suraface using density-functional-theory calculations as training data. The 2b+3b descriptor and potential module can be accesed also independently of the CTP approach. Publication is in preparation.
ahmetcik/OrthoBasis
A module for orthonormalizing an arbitrary set of one-dimensional functions within an arbitrary interval using the Gram-Schmidt process and numerical integration.
ahmetcik/CompressedSensing4MaterialsScience
A machine-learning tutorial on how to identify materials descriptors. The focus is laid on using the sure independence screening and sparsifying operator (SISSO) for regression.)