Investigates breakdown of uniqueness for corse grained representations. Uses: QM9 dataset, jax package for the derivatives
Will be cleaned up in March 2021 at the latest
Basisset information: STO-3G for OM from www.basissetexchange.org
main.py: calls functions and executes them. Should contain full workflow by Mey 2021
#derivatives and representation jax_*: group of python files that work with the python package jax or depend on it jax_representation.py: contains all representation written in this project jax_derivative.py: contains derivative functions for all representations. jax_additional_derivative.py: contains numerical derivatives and sorting functions.
jax_basis.py: contains tabular information needed to construct some of the representations. jax_math.py: contains functions used to calculate the representations.
numerical_derivative.py: numerical derivatives without need for jax, requires unsorted representations
representation_ZRN.py: unsorted representations for numerical differentiation and hashed for kernel learning
#kernel learning kernel_learning : contains my own functions for constructing a gaussian kernel, but very slow compared to qml kernel_easy_process : simplifies learning run, for hyperparameter findin
#code for preparing data for graphs plot_*: files that contain plotting functions_ plot_derivative.py: contains plotting and sorting functions for compounds, representations, ect. plot_kernel.py: contains plotting and sorting functions for kernel learning
#example files
trial_*: group of code examples_
trial_allCMderivatives.py : prints out derivatives of H2O to console
trial_calculate_time_for_rep.py : prints out calculation time of representations based on compound file
trial_readxyz_store.py : reads xyz files and stores them as compound class objects
trial_prep_for_kernel: prepares pickled files with Kernel_Result class instances for fast Machine Learning
trial_qml_kernel: uses pickled files from trial_prep_for_kernel to either draw scatter plots for analysis of errors
or to calculate new learning curve results
ect.
#image files: all stored in Images folder im_
Hints
submit multiple jobs (e.g. when running trial_full_analytical_derivatives.py) via the following commands: $ nohup python3 -u trial_full_analytical_derivatives.py 100 200 > job1.out &