List
- log2(a): return log2(x)
- sort_count(arr2): sort and unique and count array.
- entropy(x): return array of entropy.
- feature_evalue(): start the calculate entropy function.
log2(a) return the log2(x)
log2(2);
//1
log2(4);
//2
log2(6);
//2.584962500721156 //function log2
function log2(a){
var b=Math.log(a)/Math.log(2);
return b;
}Input is array. Return sort and unique array and the count of each elements.
//Input like :
a=[1,1,1,2,2,3,4,5,5,5,6,6,6,7,7,8,8];
//Output like:
b=sort_count(a);
[[1, 2, 3, 4, 5, 6, 7, 8],[3, 2, 1, 1, 3, 3, 2, 2]]
[1, 2, 3, 4, 5, 6, 7, 8]
[3, 2, 1, 1, 3, 3, 2, 2] // Use array to generate two unique array, and each count.
function sort_count(arr2){
arr=JSON.parse(JSON.stringify(arr2));
if (arr.length === 0) return arr;
arr = arr.sort(function (a, b) { return a*1 - b*1; });
var ret = [arr[0]];
for (var i = 1; i < arr.length; i++) { //Start loop at 1: arr[0] can never be a duplicate
if (arr[i-1] !== arr[i]) {
ret.push(arr[i]);
}
}
var out=[];
for(var j=0;j<ret.length;j++)
{
var count = 0;
for(let i=0; i<arr.length; i++)
{
if (arr[i]==ret[j]){
count++;
}
}
out.push(count);
}
return [ret,out];
}c
(3) [Array(4), Array(4), Array(4)]
0: (4) [1, 2, 3, 4]
1: (4) [1, 2, 3, 4]
2: (4) [1, 2, 3, 4]
d=transforme_array(c);
[Array(3), Array(3), Array(3), Array(3)]
0: (3) [1, 1, 1]
1: (3) [2, 2, 2]
2: (3) [3, 3, 3]
3: (3) [4, 4, 4]c=[[1,2,3,4],[1,2,3,4],[1,2,3,4]];
function transforme_array(a2)
{
//Deep copy array.
a=JSON.parse(JSON.stringify(a2));
//Create by feature array
var b=[];
for(var j=0;j<a[0].length;j++)
{
b.push([]);
}
// put feature into each array.
for(var i=0;i<a.length;i++)
{
for(var j=0;j<a[i].length;j++)
{
b[j].push(a[i][j]);
}
}
return b;
}Return array of entropy.
a=[1,1,1,2,2,3,4,5,5,5,6,6,6,7,7,8,8];
b=sort_count(a)
console.log(b);
// [1, 2, 3, 4, 5, 6, 7, 8],
// [3, 2, 1, 1, 3, 3, 2, 2]
console.log(entropy(b[1]));function entropy(arr2){
arr=JSON.parse(JSON.stringify(arr2));
var sum=0;
for(var i=0;i<arr.length;i++)
{
sum+=arr[i];
}
var entropy_value=0;
var array=[];
for(var i=0;i<arr.length;i++)
{
var a=arr[i]/sum;
var b=a*log2(a);
entropy_value+=b*(-1);
array.push(b);
}
return [sum,entropy_value,array];
}get data from input.
function feature_evalue()
{
//get data;
data= getinforvalue('training_decision');
var label=getinforlabel('label_decision');
var label_unique=sort_unique(label);
//
var out=sort_count(label);
console.log(out);
console.log(entropy(out[1]));
feature_array=transforme_array(data);
feature_cutoff_array=[];
for(var i=0;i<feature_array.length;i++)
{
feature_cutoff_array.push([]);
feature_cutoff_array[i]=cutoff(feature_array[i]);
}
feature_label=[];
for(var i=0;i<feature_array.length;i++)
{
feature_label.push([]);
feature_label[i]=mljs_determin_label(feature_array[i],feature_cutoff_array[i]);
}
feature_auc=[];
feature_entropy=[];
for(var i=0;i<feature_array.length;i++)
{
feature_auc.push([]);
feature_entropy.push([]);
feature_auc[i]=mljs_label_cross(label,feature_label[i]);
feature_entropy[i]=mljs_label_entropy(label,feature_label[i])
}
console.log(feature_entropy);
table_html='<table class="table" id="feature_evalue_table"> \
<thead> \
<tr> \
<th scope="col">Number</th> \
<th scope="col">AUC</th> \
<th scope="col">Information Gain</th> \
<th scope="col">Gain ratio</th> \
<th scope="col">Gini index</th> \
</tr> \
</thead> \
<tbody>';
for(var i=0;i<feature_entropy.length;i++)
{
table_html+='<tr> \
<th scope="row">'+feature_name_array[i]+'</th> \
<td>'+ feature_entropy[i][0]+'</td> \
<td>'+ feature_entropy[i][1]+'</td> \
<td>'+ feature_entropy[i][2]+'</td> \
<td>'+ feature_entropy[i][3]+'</td> \
</tr>';
}
table_html+='</tbody></table>';
document.getElementById("feature_performance").style.display = "block";
document.getElementById("feature_performance_table").innerHTML = table_html;
$('#feature_evalue_table').DataTable();
}- input is the a is the label of fact.
- b is the array of use probability to predict label.
- Return the 4 value.
- auc: the bigger the better.
- Gain: the bigger the better.
- Gain ratio: the bigger the better.
- Gini index: the smaller the better.
function mljs_label_entropy(a,b)
{
//a is the label of fact.
//b is the array of use probability to predict label.
//c is the begin entropy.
var pro=[];
var tprarr=[];
var fprarr=[];
var entropy_array=[];
var entropy_change=[];
var information_gain=[];
var other=[];
var gain_ratio_array=[];
var iv_array=[];
var gini_index_array=[];
for(var i=0;i<b.length;i++){
// each cut off label is b[i]
var tp=0;
var tn=0;
var fp=0;
var fn=0;
for(var j=0;j<b[i].length;j++){
if(b[i][j]==1 & a[j]==1){
tp++;
}
if(b[i][j]==0 & a[j]==0){
tn++;
}
if(b[i][j]==0 & a[j]==1){
fn++;
}
if(b[i][j]==1 & a[j]==0){
fp++;
}
}
//
tpr=tp/(tp+fn);
fpr=fp/(tn+fp);
//
fprarr.push(fpr);
tprarr.push(tpr);
//the total number;
var all=tp+tn+fp+fn;
// the actual number;
var postive=tp+fn;
var negative=tn+fp;
//the above the throeshold. and below the thoreshold.
var above=tp+fp;
var below=tn+fn;
// the actuall positive portion of the total number.
var pos_ratio=postive/all;
var neg_ratio=negative/all;
var tp_above=tp/above;
var fp_above=fp/above;
var tn_below=tn/below;
var fn_below=fn/below;
//gini index begin
var gini_index=(1-Math.pow((tp/above),2)-Math.pow((fp/above),2))*(above/all)+(1-Math.pow((tn/below),2)-Math.pow((fn/below),2))*(below/all);
if(isNaN(gini_index)){
gini_index=10;
}
gini_index_array.push(gini_index);
//gini index end
//iv
var iv=(-1)*(above/all)*log2(above/all)+(-1)*(below/all)*log2(below/all);
//entropy
var entropy=(-1)*(pos_ratio*log2(pos_ratio)+neg_ratio*log2(neg_ratio));
entropy_array.push(entropy);
//the entropy of classification.
var change=(-1)*(above/all)*(tp_above*log2(tp_above)+fp_above*log2(fp_above))+(-1)*(below/all)*(tn_below*log2(tn_below)+fn_below*log2(fn_below));
entropy_change.push(change);
// is the change is NaN then gain of information is 0.
var information=0;
if(isNaN(entropy-change)){
information=0;
}else{
information=entropy-change;
}
information_gain.push(information);
if(isNaN(iv)){
iv=1;
}
//
iv_array.push(iv);
var gain_ratio=information/iv;
gain_ratio_array.push(gain_ratio);
other.push(iv);
}
var auc=0;
for(var i=1;i<fprarr.length;i++){
auc+=(tprarr[i]+tprarr[i-1])*(fprarr[i]-fprarr[i-1])/2;
}
//auc=auc.toFixed(3);
//auc=Math.abs(auc).toFixed(3);
auc=Math.abs(auc);
if(auc<0.5)
{
auc=1-auc;
}
auc=auc.toFixed(4)*1;
var information_gain_value=Math.max(...information_gain).toFixed(4)*1;
var gain=Math.max(...gain_ratio_array).toFixed(4)*1;
var gini_index_min=Math.min(...gini_index_array).toFixed(4)*1;
return [auc,information_gain_value,gain,gini_index_min];
}