An experimental project under Bayesian neural networks using Langevin-gradients parallel tempering MCMC [Chandra et al,2019] which could be implemented in a parallel computing environment.
The proposal here is to compare our stock price forecasting model with state-of-art neural network training algorithms (FNN-SGD and FNN-Adam)
-
data.py - This file is used for data preprocessing.
-
nn.py - To run the results, desired parameters should be set in this file
Following are some example results of MMM’s stock price prediction. They are They are one-step, two-step, five-step prediction result and error analysis respectively. The grey area is the uncertainty of the prediction results.
- Chandra R , Jain K , Deo R V , et al. Langevin-gradient parallel tempering for Bayesian neural learning[J]. Neurocomputing, 2019, 359(SEP.24):315-326.
When you use Bayesian neural network with Parallel Tempering MCMC, please cite the above papers.