logistic_regression: Logistic Regression S3 Object

In MauroCE/LogisticRegression:

Description

Logistic Regression S3 Object

Usage

logistic_regression(
  X,
  y,
  cost = "MLE",
  method = "BFGS",
  sigmab = 1,
  niter = 100,
  alpha = 0.1,
  gamma = 0.001
)

Arguments

X	Matrix of training examples of dimensions (number of obs, number of features + 1). The first column must be a column of 1s to fit the intercept.
y	Column vector of 0-1 training labels of dimension (number of obs, 1).
cost	String indicating which cost function to optimize. Options are "MLE" or "MAP". If "MAP" is chosen an isotropic gaussian centered at zero and with 'sigmab^2 * diag(ncol(X))' as its variance-covariance matrix is placed as a prior on the coefficients. This corresponds to Ridge regularization on **all** the coefficients, including the intercept.
method	String indicating the optimization method used to optimize 'cost'. If method is 'BFGS' then the function 'optim()' is used. Otherwise, class methods are implemented for 'grad_ascent()' (performing gradient ascent) and 'newton_method()'. In case they are implemented both for the 'MLE' case and the 'MAP' case.
sigmab	Standard deviation of the univariate gaussian distribution placed on each coordinate of the vector of coefficients. It's the inverse of the regularization parameter. Should not be zero.
niter	Number of iterations that the optimization algorithm should perform. This is passed only to 'grad_ascent()' and 'newton_method()', but not to the 'optim()' function.
alpha	Learning rate for 'newton_method()'. Used to dump or enhance learning to avoid missing or not reaching the optimal solution. Could be merged with 'gamma' but defaults are different.
gamma	Learning rate for 'grad_ascent()', used to dump or enhance learning to avoid missing or or not reaching the optimal solution.

MauroCE/LogisticRegression documentation built on Jan. 18, 2020, 10:04 a.m.