nicoleyghu/chbe6746_project

ChBE6746 group 10 project

ChBE 6746 - Group 10 Project

N. Hu 2021/04/19

Package AMPtorch installation instructions

1. In a designated folder, clone the active branch of AMPtorch from github in terminal. (Using macOS/Unix syntax here. Windows may have slightly different syntax for cmd or powershell.)

   Create a folder for the cloned package under root directory (or a specified directory): mkdir ~/amptorch_MCSH_paper1; cd ~/amptorch_MCSH_paper1

   Git clone the package with specific branch:git clone --branch MCSH_paper1 https://github.com/nicoleyghu/amptorch.git

2. Navigate into the cloned folder by cd amptorch and install the package following the instructions for AMPtorch installation as stated in README.md.

   Update Conda: conda update conda

   Install environment for CPU machines: conda env create -f env_cpu.yml

   Activate the environment by: conda activate amptorch_MCSH

   Install the package with: pip install -e .

3. After Installing the package, clone the provided dataset and training python script through another github repository:

   Create a folder for the cloned package under root directory (or a specified directory): mkdir ~/chbe6746_project; cd ~/chbe6746_project

   Git clone the package with specific branch:git clone https://github.com/nicoleyghu/chbe6746_project.git

4. Go to the cloned folder and execute the train.py file to build the neural network potential for the dataset stored in ./data/

   Move to the folder: cd chbe6746_project

   Make sure the virtual environment amptorch_MCSH is active. If not, conda activate amptorch_MCSH

   Run the training script by python train.py

Problem 1 - Geometry optimization

Located in optimization_method folder.

Algorithms for geometry optimization:

• Self-implemented GA algorithm on 2D continuous space: optimize_by_ga.py by N. Hu
• Self-implemented direct search algorithm on 2D continuous space: optimize_by_positive_span.py by J. Pederson
• Baseline Scipy nelder-mead on 2D continuous space: optimize_by_nelder_mead.py by N. Hu

Problem 2 - Hyperparameter optimization

Located in hyperparameter_tuning folder.

Algorithms for hyperparameter optimization:

• Self-implemented GA algorithm on continuous and discrete space: hp_opt_by_ga.py by N. Hu
• Self-implemented TABU search algorithm on continuous and discrete space: hp_opt_by_ta.py by J. Pederson
• Baseline Scipy nelder-mead on continuous and discrete space: hp_opt_by_nelder_mead.py by N. Hu
• Baseline hyperopt on continuous and discrete space: baseline_hyperopt.py by N. Hu