# Available models
All additionl data can be found [here](https://sid.erda.dk/sharelink/EzFLHhjSI3)

| Folder | Description |
| :------- |:---------|
| full_models | Contains three folders describing which dataset ML models are trained on. All ML models are trained with best model LGBM (Opt) and all features. |
| No_duplicate_substrates | Contains all ML models from the random search |
| top2_ir_precatalyst | Contains all ML models from the ablation study |
| pka | Contains ML models to predict C-H pKa values described in [pKalculator: A pKa predictor for C–H bonds](http://dx.doi.org/10.3762/bjoc.20.144)

## Details for full models:
| dataset | Size | Training size | Reaction sites | Non-reacting sites |
| :------- |:---------|:---------|:---------|:---------|
| literature_regio_yield_above_30 | 788 | 4633 | 788 | 3845 |
| no_duplicate_substrates | 502 | 2900 | 502 | 2398 |
| top2_ir_precatalyst | 667 | 4034 | 667 | 3367 |

## Details for No_duplicate_substrates
- training size: 489 compounds, 2861 sites
- test size : 163 compounds,  925 sites

## Details for top2_ir_precatalyst
- training size: 533 compounds, 3246 sites
- test size : 186 compounds,  788 sites