class ExternalMaxentOptimizer implements MaxentOptimizerInterface
This class enables the use of a program written in a different language to optimize our model and return the weights for use in php.
Mostly common use: Optimize in a fast compiled language (ex.: C) or
a language with great libraries (ex.: Matlab)
Output a json array that contains all the necessary information to
train the model and determine the weights that maximize the
conditional log likelihood of the training data
The array has one element for each training document. The element is
a map with each possible class as key and an array of features that
fire for each class. There is a special key called '__label__' that
contains a string value which is the actual class of this document
Ex.:
[
{
"class1": ["feature1","feature2","feature3"],
"class2": ["feature1","feature4","feature5"],
"__label__": "class1"
},
{
"class1": ["feature2","feature3"],
"class2": ["feature1","feature4","feature5"],
"__label__": "class1"
},
{
"class1": ["feature1","feature2","feature3"],
"class2": ["feature1"],
"__label__": "class2"
}
]
Send this array to an external program that will return a map of
floats in json that will contain the weight for each feature.
Methods
| __construct(string $optimizer) | ||
| array |
optimize(array $feature_array)
This function receives an array that contains an array for each document which contains an array of feature identifiers for each class and at the special key '__label__' the actual class of the training document. |
Details
at line 51
public
__construct(string $optimizer)
at line 63
public array
optimize(array $feature_array)
This function receives an array that contains an array for each document which contains an array of feature identifiers for each class and at the special key '__label__' the actual class of the training document.
As a result it contains all the information needed to train a
set of weights with any target. Ex.: If we were training a maxent
model we would try to maximize the CLogLik that can be calculated
from this array.