mcordingley / regression Goto Github PK

I'm a self-taught beginner in statistics and I'm still wrapping my head around the concepts as I go, so please feel free to correct anything I say.

With that said, I would like to be able to perform a linear regression on some data and than calculate the statistical significance of the regression based on the correlation coefficient and the number of data points.

Basically I want to know

• the direction of the trend (the slope of the line from linear regression)
• the strength of the relationship (the coefficient of determination, aka R-Squared)
• and the significance of the relationship, calculated with a pretty complicated formula, or read from a table:

See http://www.oneonta.edu/faculty/vomsaaw/w/psy220/files/SignifOfCorrelations.htm

How would you suggest I go about getting the significance of the correlation?

Is it something that makes sense to include in this library? Should I include it in a wrapper, a level above?

(For my own future reference
http://vassarstats.net/rsig.html - calculates significance from r and N, but the code in JS is ugly
http://janda.org/c10/Lectures/topic06/L24-significanceR.htm - more notes about strength and significance)

Looking at the math, it seems possible to add regularization to the gradient descent component. This would avoid over-fitting by penalizing a fit for having larger or more coefficients. Should be able to build it as a decorator over the gradient objects.

Descent regression either deconverges or simply comes to rest on bad values.

What do you think about following the internal PHP-FIG conventions and naming all interfaces ..Interface?
I would also move the interfaces one level deeper, LinkingInterface into mcordingley\Regression\Linking and RegressionStrategyInterface into mcordingley\Regression\RegressionStrategy

See https://github.com/php-fig/fig-standards/blob/master/bylaws/002-psr-naming-conventions.md

It would make the code just a little bit more readable.

This includes things like logarithmic, exponential, and power regression.

Need to verify the math.