import get from 'lodash-es/get.js'
import each from 'lodash-es/each.js'
import flatten from 'lodash-es/flatten.js'
import range from 'lodash-es/range.js'
import isbol from 'wsemi/src/isbol.mjs'
import iseobj from 'wsemi/src/iseobj.mjs'
import isearr from 'wsemi/src/isearr.mjs'
import MLR from 'ml-regression-multivariate-linear'
/**
* 針對矩陣X與矩陣Y數據進行多變數線性回歸，單1種y時(y=b+m1*x1+m2*x2+...)，多種y時[ (y1=b1+m11*x1+m12*x2+...), (y2=b2+m21*x1+m22*x2+...),... ]
*
* Unit Test: {@link https://github.com/yuda-lyu/w-statistic/blob/master/test/regMpLine.test.js Github}
* @memberOf w-statistic
* @param {Array} matX 輸入X二維陣列，為[ [x11,x12,x13,...], [x21,x22,x23,...],... ]
* @param {Array} matY 輸入Y二維陣列，為[ [y11,y12,...], [y21,y22,...],... ]
* @param {Object} [opt={}] 輸入設定物件，預設{}
* @param {Array} [opt.interpX=null] 輸入經由回歸結果內插指定x陣列，為[x1,x2,x3,...]，預設null
* @param {Boolean} [opt.useRegIntercept=true] 輸入是否回歸使用截距布林值，預設true
* @param {Boolean} [opt.useSync=false] 輸入是否使用同步函數布林值，預設false
* @returns {Object|Promise} 若useSync=true回傳回歸結果物件，若useSync=false則回傳Promise，此時若成功則resolve回歸結果物件，若失敗則reject錯誤訊息
* @example
*
* async function test() {
*
* let arrX
* let arrY
* let r
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = await regMpLine(arrX, arrY)
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515 ],
* // [ -1.4210854715202004e-14 ],
* // [ 3.552713678800501e-15 ]
* // ],
* // m1: 2.0000000000000515,
* // m2: -1.4210854715202004e-14,
* // b: 3.552713678800501e-15
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0, 0],
* [2, 4],
* [4, 6],
* [6, 8],
* ]
* r = await regMpLine(arrX, arrY)
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515, 3.552713678800501e-14 ],
* // [ -1.4210854715202004e-14, 1.9999999999999716 ],
* // [ 3.552713678800501e-15, 0 ]
* // ],
* // m11: 2.0000000000000515,
* // m21: 3.552713678800501e-14,
* // m12: -1.4210854715202004e-14,
* // m22: 1.9999999999999716,
* // b1: 3.552713678800501e-15,
* // b2: 0
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0, 0, 0],
* [2, 4, 3],
* [4, 6, 5],
* [6, 8, 7],
* ]
* r = await regMpLine(arrX, arrY)
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515, 3.552713678800501e-14, 1.0000000000000426 ],
* // [ -1.4210854715202004e-14, 1.9999999999999716, 0.9999999999999591 ],
* // [ 3.552713678800501e-15, 0, 7.105427357601002e-15 ]
* // ],
* // m11: 2.0000000000000515,
* // m21: 3.552713678800501e-14,
* // m31: 1.0000000000000426,
* // m12: -1.4210854715202004e-14,
* // m22: 1.9999999999999716,
* // m32: 0.9999999999999591,
* // b1: 3.552713678800501e-15,
* // b2: 0,
* // b3: 7.105427357601002e-15
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = await regMpLine(arrX, arrY, { useRegIntercept: false }) //不使用截距, 也就是截距b=0
* console.log(r)
* // => {
* // coes: [ [ 1.9999999999999716 ], [ 1.4210854715202004e-14 ] ],
* // m1: 1.9999999999999716,
* // m2: 1.4210854715202004e-14,
* // b: 0
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = await regMpLine(arrX, arrY, { interpX: [0, 0] })
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515 ],
* // [ -1.4210854715202004e-14 ],
* // [ 3.552713678800501e-15 ]
* // ],
* // m1: 2.0000000000000515,
* // m2: -1.4210854715202004e-14,
* // b: 3.552713678800501e-15,
* // interpX: [ 0, 0 ],
* // interpY: [ 3.552713678800501e-15 ]
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = await regMpLine(arrX, arrY, { interpX: [100, 0] })
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515 ],
* // [ -1.4210854715202004e-14 ],
* // [ 3.552713678800501e-15 ]
* // ],
* // m1: 2.0000000000000515,
* // m2: -1.4210854715202004e-14,
* // b: 3.552713678800501e-15,
* // interpX: [ 100, 0 ],
* // interpY: [ 200.00000000000514 ]
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = await regMpLine(arrX, arrY, { interpX: [0, 100] })
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515 ],
* // [ -1.4210854715202004e-14 ],
* // [ 3.552713678800501e-15 ]
* // ],
* // m1: 2.0000000000000515,
* // m2: -1.4210854715202004e-14,
* // b: 3.552713678800501e-15,
* // interpX: [ 0, 100 ],
* // interpY: [ -1.4175327578413999e-12 ]
* // }
*
* arrX = [
* [0, 0],
* [1, 2],
* [2, 3],
* [3, 4],
* ]
* arrY = [
* [0],
* [2],
* [4],
* [6],
* ]
* r = regMpLine(arrX, arrY, { useSync: true }) //使用同步函數(sync)
* console.log(r)
* // => {
* // coes: [
* // [ 2.0000000000000515 ],
* // [ -1.4210854715202004e-14 ],
* // [ 3.552713678800501e-15 ]
* // ],
* // m1: 2.0000000000000515,
* // m2: -1.4210854715202004e-14,
* // b: 3.552713678800501e-15
* // }
*
* }
* test()
* .catch((err) => {
* console.log(err)
* })
*
*/
function regMpLine(matX, matY, opt = {}) {
//interpX
let interpX = get(opt, 'interpX')
if (!isearr(interpX)) {
interpX = null
}
//useRegIntercept, 是否回歸使用截距
let useRegIntercept = get(opt, 'useRegIntercept')
if (!isbol(useRegIntercept)) {
useRegIntercept = true
}
//useSync
let useSync = get(opt, 'useSync')
if (!isbol(useSync)) {
useSync = false
}
//_sync
let _sync = () => {
//check matX
if (!isearr(matX)) {
throw new Error(`matX is not an effective array`)
}
let matX0 = get(matX, 0, [])
if (!isearr(matX0)) {
throw new Error(`matX[0] is not an effective array`)
}
//check matY
if (!isearr(matY)) {
throw new Error(`matY is not an effective array`)
}
let matY0 = get(matY, 0, [])
if (!isearr(matY0)) {
throw new Error(`matY[0] is not an effective array`)
}
//MLR
let optMlr = {
intercept: useRegIntercept,
}
let regression = new MLR(matX, matY, optMlr)
// console.log('regression', regression)
// console.log(regression.predict(X)) // Apply the model to X
// weights: [
// [ 0.4367273586800948 ],
// [ 1.4356527329046394 ],
// [ -0.10327272210504645 ]
// ],
//r
let r = {
coes: regression.weights,
}
if (useRegIntercept) {
if (regression.outputs === 1) {
let vs = flatten(regression.weights)
each(vs, (v, ir) => {
if (ir === regression.inputs) {
r['b'] = v
}
else {
r[`m${ir + 1}`] = v
}
})
}
else {
each(range(regression.inputs + 1), (ir) => {
each(range(regression.outputs), (ic) => {
let v = regression.weights[ir][ic]
if (ir === regression.inputs) {
r[`b${ic + 1}`] = v
}
else {
r[`m${ic + 1}${ir + 1}`] = v
}
})
})
}
}
else {
if (regression.outputs === 1) {
let vs = flatten(regression.weights)
each(vs, (v, ir) => {
r[`m${ir + 1}`] = v
})
r['b'] = 0
}
else {
each(range(regression.inputs + 1), (ir) => {
each(range(regression.outputs), (ic) => {
let v = regression.weights[ir][ic]
r[`m${ic + 1}${ir + 1}`] = v
if (ir === regression.inputs) {
r[`b${ic + 1}`] = 0
}
})
})
}
}
// console.log('r', r)
//interpX
if (isearr(interpX)) {
let interpY = regression.predict(interpX)
r.interpX = interpX
r.interpY = interpY
}
return r
}
//_async
let _async = async () => {
let r = null
try {
r = _sync()
if (iseobj(r)) {
return r
}
else {
return Promise.reject(`no effective data`)
}
}
catch (err) {
console.log(err)
return Promise.reject(err.message)
}
}
if (useSync) {
return _sync()
}
else {
return _async()
}
}
export default regMpLine