This repository contains recurrent neural networks for polymeric property prediction as described in the paper Dielectric Polymer Property Prediction Using Recurrent Neural Networks with Optimizations. Specific optimization techniques that are critical for achieving high learning speed and accuracy were developed. Together with the compact binary and non binary representations of SMILES fingerprints, modification of the back propagation learning was performed based on the affine transformation of the input sequence (ATransformedBP) as well as the resilient backpropagation (iRPROP-) was improved with initial weight update parameter optimizations. Both AtransformedBP and iRPROP- with optimization models are trained on a single-tasking.