This repo contains the code for the ML competition held at drivendata.com
The objective of this competition was to forecast energy consumption from varying amounts of "cold start" data, and little other building information. That means that for each building in the test set you are given a small amount of data and then asked to predict into the future.
Why LSTM? What is it? I wanted to get hands on experience with Deep Learning and LSTM seemed like a possible solution for this problem as recurrent networks are quite well suited for forecasting of sequential data (time series in this case). LSTM (Long Short Term Memory) Networks are advance version of recurrent neural networks with some additional features. In LSTM, the network is capable of learning and maintaining a memory overtime while showing gradual improvement.
I was in top 10% of the competition. Here's the rank
