Bayesian LSTM Implementation in PyTorch
Inspired by: Deep and Confident Prediction for Time Series at Uber (2007) https://arxiv.org/pdf/1709.01907.pdf
Bayesian Neural Networks are gaining interest due to their highly desirable properties of providing quantifiable uncertainties and confidence intervals, unlike equivalent frequentist methods.
This repository demonstrates an implementation in PyTorch and summarizes several key features of Bayesian LSTM (Long Short-Term Memory) networks through a real-world example of forecasting building energy consumption. The Appliances energy prediction dataset used in this example is from the UCI Machine Learning Repository (https://archive.ics.uci.edu/ml/datasets/Appliances+energy+prediction)
The accompanying notebook is shared directly from Google Colab. As a result, interactive visualizations have not been transferred to GitHub.
Please view the notebook in Google Colab by clicking the Open in Colab button or by clicking here: https://colab.research.google.com/drive/1pwMzsdRPwTO8oRVU0LwY9hs_z-ye67su