We present the dataset and python scripts used in our autoencoder (AE) neural network (NN)-based reduced chemistry work (Zhang and Sankaran, 2022). 1. To train the AE NN, run "python Train_PSR_AE_PCA.py " with Keras 2. The dataset is about syngas combustion inside 0-D steady perfectly stirred reactor (PSR) at a wide range of parameter conditions. The fuel is CO, H2, N2 with a volume ratio 5:1:4. The oxidizer is O2 and N2 mixed in 1:3 by volume. The inflow temperature is 500 K and combustion occurs at atmospheric pressure. In total, there are 1.63 million samples with equivalence ratio varying from 0.09 to 20.0 and residence time scale varying to cover the entire S-curve. The dataset is in hdf5 format and can be loaded with the python script, load_data_h5.py. Inside the dataset, there 12 entries. 1. asciiListtmp = h5f['vars_name'][()] ##name of the 12 thermochemical state variables 2. para_Phi = h5f['parameters_Phi'][()] ##equivalence ratio, varying from 0.09 to 20.0 3. para_Tin = h5f['parameters_Tin'][()] ##inflow temperature, constant=500 [K] 4. para_inv_tau_res = h5f['parameters_inv_tau_res'][()] ##inverse of residence time, varying from 4.53e-09 to 1.54e+04 [1/s] 5. x_train_min = h5f['trainset_min'][()] ##minimum value of training set 6. x_train_max = h5f['trainset_max'][()] ##maximum value of training set 7. x_data = h5f['dataset'][()] ##Thermochemical state variables (temperature, mass fractions of chemical species), normalized with x_train_min and x_train_max to be [-1,1] 8. x_src_data = h5f['dataset_src'][()] ##source term * 2/(x_train_max-x_train_min) 9. x_del_data = h5f['dataset_del'][()] ##(xinflow-x)* 2/(x_train_max-x_train_min) 10. train_ind = h5f['train_dataset_indices'][()] #0.7, sample index of training set 11. test_ind = h5f['test_dataset_indices'][()] #0.3*0.5, sample index of test set 12. vali_ind = h5f['valid_dataset_indices'][()] #0.3*0.5, sample index of validation set The training/test/validation splitting is used in our reduced chemistry work.