Safe Check

Safe Check

Safe checks are implemented in each iteration for checking the validity of estimates. Some of the following check methods should be able to be turned off by the user.

• Start value (implemented)
  Start value is checked at the beginning. η is calculated by the first observation and the start value of θ. Whether η is in the support is checked. Stop if η is out of the support.

• Number of observations (implemented)
  The number of observations is checked at the beginning. The number of observations is compared to the rank of X. Stop if the number of observations is smaller.

• Gradient (implemented)
  The gradient is checked at each iteration. Stop is the gradient is NA or non-finite.

• Non-finite estimates (implemented)
  The estimate is checked at each iteration. Warn if a coefficient is non-finite.

• Deviance (optional) (implemented)
  The deviance is checked at each iteration. Stop if the deviance is non-finite.

• η (implemented)
  Whether η is in the support is checked at each iteration. Stop if η is not in the support.

• Variance of expectation of Y (implemented)
  Var(μ) is checked at each iteration. Stop if var(μ) is NA or 0.

• Convergence (optional) (implemented)
  Convergence is checked in the end. Warn if the change in the deviance of last two iterations is larger than the threshold.

• Boundary
  Checked in the end. Warn if heuristic methods are used.

• Validity of μ (implemented)
  Checked in the end. For binomial family, warn if μ is 0 or 1. For Poisson family, warn if μ is 0.

• Number of covariates (implemented)
  Checked in the end. If the number of covariates is greater than the rank of X, excessive covariates will be set to NA and we will throw a warning.

• Validity of estimates
  Checked in the end. Check if η is in the support for all observations.