Feature Scaling - utkaln/machine-learning GitHub Wiki

Feature Scaling is about making the features comparable to one another

• Target to keep the value range between -1 < x < 1

1. Number Divided by Maximum : x / max
2. Mean normalization : (x - mean) / (range)
3. Z-score normalization : (x - mean) / std. dev

Optimize Alpha (Learning rate)

• Too large value will miss the minimum global cost, so keep it as small as possible
• Too small value will take many many iterations to find the global minimum cost
• Start with 0.001, then try 0.003, 0.01, 0.003, 0.1, 0.3, 1