Skip to content

Latest commit

 

History

History
39 lines (30 loc) · 1.79 KB

predictor.md

File metadata and controls

39 lines (30 loc) · 1.79 KB
Raw

Predictor

1. Prepare dataset

Once you download default data by running the following command,

$ mofreinforce download default

Then, examples of dataset of predictors will be downloaded at data/dataset_predictor/qkh or data/dataset_predictor/selectivity

The dataset directory should include train.json, val.json and test.json.

The json consists of names of structures (key) and dictionary of descriptions of structures (values). The descriptions include topo_name, mc_name, ol_name, ol_selfies, topo, mc, ol, target.

(optional)

  • topo_name : (string) name of topology.
  • mc_name : (string) name of metal cluster.
  • ol_name : (string) name of organic linker. (required) The topologies, metal clusters and organic linkers should be vectorized. Given topologies and metal clusters are categorical variables, they need to be converted to idx.
  • topo : (int) index of topology. The index can be found in data/mc_to_idx.json
  • mc : (int) index of metal cluster. The index can be found in data/topo_to_idx.json When it comes to organic linkers, it is represented by SELFIES.
  • ol_selfies : (string) SELFIES string of organic linker. The smiles can be converted into SELFIES using sf.decode(SMILES) in libs/selfies.
  • ol : (list) a list of index of SELFIES string. The index of vocabulary of SELFIES can be found in data/vocab_to_idx.json Finally, the target property you want to optimize should be defined.
  • target : (float) target property.

2. Training predictor

Here is an example to train predictor for heat of adsorption in the mofreinforce directory

# python run_predictor.py with regression_qkh_round3

By modifying predictor/config_predictor.py, you can train your predictors.