com.mayabot.nlp.segment.plugins.pos.PosPerceptron.kt Maven / Gradle / Ivy
package com.mayabot.nlp.segment.plugins.pos
import com.mayabot.nlp.perceptron.PerceptronModel
import com.mayabot.nlp.perceptron.PerceptronRunner
import com.mayabot.nlp.segment.Nature
import java.io.File
/**
* 词性分析感知机
*/
class POSPerceptron(
val perceptron: PerceptronModel) {
init {
perceptron.decodeQuickMode(true)
}
val labels = InnerPos.natures
val natureList = labels.map { Nature.parse(it) }.toTypedArray()
val runner = PerceptronRunner(PosPerceptronDef(labels))
init {
//learn("[m]/x 章/q")
// [m]代表数字，但是这个转换逻辑现在在代码里面找不到了
// 而且[] 会被转换为中文书名号《》，肯定哪里错了
// 导致这个特征在FeatureSet中找不到，导致很大
}
fun decodeNature(sentence: List): List {
val result = decodeWithIndex(sentence)
return result.map {
if (it == -1) {
Nature.x
} else {
natureList[it]
}
}
}
/**
* 解码
*/
fun decodeWithIndex(list: List): IntArray {
return runner.decode(perceptron, list)
}
fun save(dir: File) {
perceptron.save(dir)
}
fun learn(sample: String) {
runner.learnModel(perceptron, sample)
}
/**
* 解码
*/
fun decode(list: List): List {
val decodeResult = runner.decode(perceptron, list)
return decodeResult.map { labels[it] }
}
companion object {
fun train(labels: List,
trainFile: File,
evaluateFile: File?,
iter: Int,
threadNum: Int): PerceptronModel {
val runner = PerceptronRunner(PosPerceptronDef(labels.toTypedArray()))
return runner.train(trainFile, evaluateFile, iter, threadNum, true)
}
}
}