Machine Learning in Elixir

Tempering Expectations

After running these small experiments on stock prices, you might feel there’s some promise in using neural networks to predict the future. After all, both of the neural networks you’ve trained do a pretty good job of following the trend line of AAPL’s stock price over the course of ten years. In reality, forecasting the future is really challenging.

Neural networks are powerful interpolators. Interpolation is the process of filling in unknown data points within a fixed domain. Interpolation is explicitly different from extrapolation because in extrapolation we attempt to fill in unknown data points from outside a fixed domain. What does this have to do with time-series analysis?

Well, you may have noticed that your neural ...