This repository presents the procedure we performed in the paper with title: "Path sampling of recurrent neural networks by incorporating known physics", where we include two jupyter notebooks showing how we sampled subset from constraint-free LSTM in a biased manner via path sampling. The subset can then be used to re-train new LSTMs to obtain a LSTM model with constrained thermodynamic or dynamic constraints. In this repository we also include the 200ns trajectory from molecular dynamics simulation of Aib9 used in the paper. The totally 10000 predicted trajectories used for sampling are too large to upload, please email the author for the data.
