Toward emulating an explicit organic chemistry mechanism with a random forest model: dataset and training code

This repository contains the dataset created with the GECKO-A model and the code (training_gecko_rf_final.py) used to train and test random forests for predicting secondary organic aerosol formation. For each simulation, results are distributed in two separate files identified as such: <precursor>_library_<id>_predictors.csv and <precursor>_library_<id>_outcomes.csv. <precursor> is either ARO1 (toluene) or dodecane_4gen (dodecane). <id> is a unique simulation identifier. the *predictors.csv files contain the state of the predictors for each timestep at the beginning of the chemical solver integration step. the *outcomes.csv files contain the state of the outcomes at the end of the chemical solver integration step. The TRAINING_* directories contain training simulations. TRAINING_ALL contains all the training data, used for the default random forest configuration. TRAINING_*NOX contain sorted training data matching LOW, MID and HIGH NOx initial regimes (see associated article) to train the specialized random forests. Similarly, the VALIDATION_* directories contain validation simulations, used to test the random forests after training. The TESTING* directories contain the results of testing the random forest for comparison with the VALIDATION simulations.