crf_caretmethod {crfsuite}

Functionality allowing to tune a crfsuite model using caret

Description

The object crf_caretmethod contains functionality to tune a crf model using caret. Each list elment of crf_caretmethod is a list of functions which can be passed on to the method argument of caret::train to tune the hyperparameters of the crfsuite model.

Usage

crf_caretmethod

Format

see details

Details

If you want to tune the hyperparameters of a crfsuite model (see crf_options and the options argument of crf), you can use the caret package.
In order to facilitate this tuning, an object called crf_caretmethod has been made available. The object crf_caretmethod is a list with 6 elements, where each of these 6 elements can be used in tuning the CRF hyperparemeters by passing it on to the method argument of the train function of the caret package.
The list has elements 'default', 'lbfgs', 'l2sgd', 'averaged_perceptron', 'passive_aggressive' and 'arow'. Each list element corresponds to arguments that you need to tune for each method as used in crf.
For crf_caretmethod

1. lbfgs: Tuning across all hyperparameters for method lbfgs: L-BFGS with L1/L2 regularization

2. l2sgd: Tuning across all hyperparameters for method l2sgd: SGD with L2-regularization

3. averaged_perceptron: Tuning across all hyperparameters for method averaged-perceptron: Averaged Perceptron

4. passive_aggressive: Tuning across all hyperparameters for method passive-aggressive: Passive Aggressive

5. arow: Tuning across all hyperparameters for method arow: Adaptive Regularization of Weights (AROW)

6. default: Tune over the hyperparameters feature.minfreq, feature.possible_states, feature.possible_transitions, max_iterations. While tuning these, it uses the default hyperparameters for each method. This tuning allows you to compare the 5 methods.

For details on the hyperparameter definitions: see crf_options