NuPIC provides a temporal memory based on sparse distributed representations. This hack attempts to encode daily historical stock market data into sequences and train NuPIC to recognize them. The key idea is to select occurrences of specific stock movements in the historical data. Based on the assumption that patterns exist in the market, train a model to recognize the lead-in sequences for the selected events. Use the anomaly score to determine if an arbitrary sequence is similar to the training sequences. If successful, a user could use the model to predict the occurrence of market events

Built With

Share this project:
×

Updates