Creates a reducer that computes a ridge regression with numX independent variables (not including constant) followed by numY dependent variables. Ridge regression is a form of Tikhonov regularization which shrinks the regression coefficients by imposing a penalty on their size. With this implementation of ridge regression there NO NEED to include a constant value for bias.
The first output is a coefficients array with dimensions (numX + 1, numY); each column contains the coefficients for the corresponding dependent variable plus the intercept for the dependent variable in the last column. Additional outputs are a vector of the root mean square of the residuals of each dependent variable and a vector of p-values for each dependent variable. Outputs are null if the system is underdetermined, e.g., the number of inputs is less than numX + 1.
| Usage | Returns | ee.Reducer.ridgeRegression(numX, numY, lambda) | Reducer |
| Argument | Type | Details | numX | Integer | the number of independent variables being regressed. |
numY | Integer, default: 1 | the number of dependent variables. |
lambda | Float, default: 0.1 | Regularization parameter. |