Feature scales

• if features are on similar scale then its easy for gradient descent to converge
  • like bedrooms(no of romms) and size (sq ft) are different scals
  • so we scale them sqft/2000 and bedrooms . = no of bedrooms/5 (mame them smaller)
  • as we get both features between 0 and 1
  • In reality we try to get features into -1 and 1 range or close enough to -1 and 1
  • sometimes we also might have to scale up extremly small features
  • sometimes we can also do mean normalization
    • replace xi with xi-ui to make features approx to 0. like if avg size = 1000 then we can do xi-1000/2000 so it is in between -1 and 1

Debug gradient descent