Neural network tutorial - [Part 2] - Why should I scale my data?
Why should I scale my data?
In this video we look at the practical implications of scaling your input and output data. If you have watched other neural network tutorials you probably noticed that it is just a habit to scale data so that the values do not deviate far from zero. Why is that exactly? What are the implications for learning the parameters of the model?
In this video I show the practical implications of scaling your input and output data and how that affects fitting.
Things shown:
1) how scaling affects the rate of learning
2) how scaling affects the initial guesses of the slope and intercept
3) and a shameless pitch about how awesome seaborn is
If you have questions about the derivatives, learning rate, or other methods of scaling data leave them in the comments.
Also check out some other helpful channels and videos on the broader topics of machine learning and cool math stuffs:
Note: Neural networks are not the best method for fitting a line there is a much more direct way for doing this which is exactly what most math libraries use. This is just to build up understand of neural networks.