Create a starting point for starting the PENSE algorithm in `pense()`

.
Multiple starting points can be created by combining starting points via
`c(starting_point_1, starting_point_2, ...)`

.

starting_point(beta, intercept, lambda, alpha) as_starting_point(object, specific = FALSE, ...) # S3 method for enpy_starting_points as_starting_point(object, specific = FALSE, ...) # S3 method for pense_fit as_starting_point(object, specific = FALSE, alpha, lambda, ...) # S3 method for pense_cvfit as_starting_point( object, specific = FALSE, alpha, lambda = c("min", "se"), se_mult = 1, ... )

beta | beta coefficients at the starting point. Can be a numeric vector, a sparse vector of class dsparseVector, or a sparse matrix of class dgCMatrix with a single column. |
---|---|

intercept | intercept coefficient at the starting point. |

lambda | optionally either a string specifying which penalty level to use
( |

alpha | optional value for the |

object | an object with estimates to use as starting points. |

specific | whether the estimates should be used as starting points only at the penalization level they are computed for. Defaults to using the estimates as starting points for all penalization levels. |

... | further arguments passed to or from other methods. |

se_mult | If |

an object of type `starting_points`

to be used as starting point for `pense()`

.

A starting points can either be *shared*, i.e., used for every penalization level PENSE
estimates are computed for, or *specific* to one penalization level.
To create a specific starting point, provide the penalization parameters `lambda`

and `alpha`

.
If `lambda`

or `alpha`

are missing, a shared starting point is created.
Shared and specific starting points can all be combined into a single list of starting points,
with `pense()`

handling them correctly.
Note that specific starting points will lead to the `lambda`

value being added to the
grid of penalization levels.
See `pense()`

for details.

Starting points computed via `enpy_initial_estimates()`

are by default *shared* starting points
but can be transformed to *specific* starting points via
`as_starting_point(..., specific = TRUE)`

.

When creating starting points from cross-validated fits, it is possible to extract only the
estimate with best CV performance (`lambda = "min"`

), or the estimate with CV performance
statistically indistinguishable from the best performance (`lambda = "se"`

).
This is determined to be the estimate with prediction performance within
`se_mult * cv_se`

from the best model.

Other functions for initial estimates:
`enpy_initial_estimates()`

,
`prinsens()`