I’ve been trying to learn more about Liquid Neural Networks and their application with respect to time series data. I’ve found the ncps library and am struggling a bit to translate their very basic sine wave example into use with time series data such as financial market prices or inventory/sales data. I’m wondering if there is anyone who might be able to provide a little more help/insight.
I’m able to create a Neural Circuit Policy and LTC model from their example, however the training and implementation with respect to a more realistic dataset is not obvious to me. I can’t seem to locate any concrete examples of this on-line and, unfortunately, the original paper while fascinating, is beyond my grasp when it comes to translating it into a pytorch model.
As a very basic example, I’m attempting to take the closing price of a security and predict its future value n days forward in time. When following some other similar examples such as using a LSTM model I am able to train and get results, but fear I’m not actually implementing the model correctly.
per the example in their documentation if I provide a time window of say 30 days, I actually get a prediction output which is also 30 days worth of data, and while I can supply a target for each day I’m really only looking for the prediction those past 30 days will provide. Simply taking the last index of the prediction doesn’t really feel to me like I’m properly using this model.
Could anyone point me in the direction of a more representative example, or be willing to connect and expand my understanding of the proper usage?