Calculate the negative log likelihood of CRF model

1 2 |

`par` |
The parameter vector of CRF |

`crf` |
The CRF |

`instances` |
The training data matrix of CRF model |

`node.fea` |
The list of node features |

`edge.fea` |
The list of edge features |

`node.ext` |
The list of extended information of node features |

`edge.ext` |
The list of extended information of edge features |

`infer.method` |
The inference method used to compute the likelihood |

`...` |
Other parameters need by the inference method |

This function calculates the negative log likelihood of CRF model as well as the gradient. This function is intended to be called by optimization algorithm in training process.

In the training data matrix `instances`

, each row is an instance and
each column corresponds a node in CRF.
The variables `node.fea`

, `edge.fea`

, `node.ext`

, `edge.ext`

are lists of length equal to the number of instances, and their elements are
defined as in `crf.update`

respectively.

This function will return the value of CRF negative log-likelihood.

`crf.update`

, `train.crf`

Questions? Problems? Suggestions? Tweet to @rdrrHQ or email at ian@mutexlabs.com.

All documentation is copyright its authors; we didn't write any of that.