Building a RNN-LSTM completely from scratch (no libraries!)

In this post, we are going to build a RNN-LSTM completely from scratch only by using numpy (coding like it’s 1999). LSTMs belong to the family of recurrent neural networks which are very usefull for learning sequential data as texts, time series or video data. While traditional feedforward networks consist of an input layer, a hidden layer, an output layer and the weights, bias, and activation function between each layer, the RNN network incorporates a hidden state layer which is connected to itself (recurrent).

Forecasting the Bitcoin price using data from Twitter and Reddit

1 Comment

In this blog post, we are going to forecast the Bitcoin price based on text data from Twitter and Reddit. Given that the observed Bitcoin price is formed by some supply and demand function, modeling the demand side, while assuming that the supply side behaves somehow stable, we may end up with some outstanding forecasting results. Social media data has been massively used in the financial industry and requires algorithms that can scale. However, social media data is unstructured and noisy. Supervised learning techniques are strongly domain dependent and need a massive amount of labeled data to be trained on to generalize well. We are going to tackle this problem by mapping the vectorized text data and sentiment directly to future price movements of Bitcoin. The economic theory claims that the price of an asset is a composition of its utility and speculation value. In 2017, we observed a crypto-currency market that went skyrocket – in the absence of a blockchain killer application so far; it is safe to assume that the reason behind this was driven by at least 90% of speculation and 10% by the utility. This assumption highly encourages our project.