Dflow_high throughput screening_machine learning

[ZijianCai]

dflow-MP_cgcnn_jupyter

https://github.com/ZijianCai/dflow/tree/main/MP_cgcnn_jupyter This is a simple introduction to dflow, materials screening and machine learning.

Setup

1. Setup conda environment (assume you have conda installed)

   conda create -n dflow-helloworld python=3.9

2. Activate conda environment and install jupyter notebook

   conda activate dflow-jupyter
   pip install notebook

How to use

1. Clone this repository to local

   git clone https://github.com/git clone https://github.com/

2. cd into the repo

   cd dflow_jupyter

3. Start up jupyter notebook

   jupyter notebook

4. Docker images should include python, pymatgen, PyTorch and scikit-learn.

Prepare your screening code

1. Get your API key from materials project: https://legacy.materialsproject.org/open. Use it as your first argument (sys.argv[1])

2. Write your criteria for screening. In the current version, the number of elements (nelements) are applied to screen as a demo. For your onwer purpose, you can write different screening criteria, such as must include certain elements, energy above hull must be smaller than a certain value, band gap must be smaller than a certain value, et al.

3. Note: the screening code will output a csv file (id_prop.csv) to the designed folder (sys.argv[2])

Prepare CGCNN code

1. The CGCNN code has been prepared, which is from: https://github.com/txie-93/cgcnn