Data Standardization (rsigley?)

Collapse
X
 
  • Time
  • Show
Clear All
new posts
  • 339955
    Restricted User
    • 07-20-12
    • 198

    #1
    Data Standardization (rsigley?)
    If you are doing a linear regression on a single variable, will you standardize the variable by doing (x-u)/stdv? When is this good and when is this not good?

    I just took some data and did a regression using descent gradients to predict points scored based off of minutes played by players on a team. I found my regression had about 7% more predictive power when I did it based off of minutes played standardized. The predictions based on the data, and then the standardized data were very different though from each other.

    any thoughts and help please!
  • HUY
    SBR Sharp
    • 04-29-09
    • 253

    #2
    Originally posted by 339955
    If you are doing a linear regression on a single variable, will you standardize the variable by doing (x-u)/stdv? When is this good and when is this not good?

    I just took some data and did a regression using descent gradients to predict points scored based off of minutes played by players on a team. I found my regression had about 7% more predictive power when I did it based off of minutes played standardized. The predictions based on the data, and then the standardized data were very different though from each other.

    any thoughts and help please!
    Linear regression is an analytically solved problem, assuming the loss function is squared loss, so why would you do gradient descent?
    Comment
    SBR Contests
    Collapse
    Top-Rated US Sportsbooks
    Collapse
    Working...