I am working with neural networks and need to deal with cyclical values, such as time of day or angles. What are the best approaches or techniques for representing and processing cyclical data to ensure accurate predictions and learning?
1 Like
For cyclical data like time of day or angles, think of it like a clock. Instead of treating 23:59 and 00:01 as far apart, wrap them around. One fun trick is to use sine and cosine functions. They can help capture that circular vibe. For example, convert time to radians and then use sin and cos to get your features. This way, the model gets the whole picture and not just a straight line. I’ve used this for predicting when people are most active online. Worked like a charm😄
1 Like