public static MnistModel ReadImage(string path, bool normalize, bool trainAsEncoder)
{
var bitmap = new Bitmap(path);
var fileName = path.Substring(path.LastIndexOfAny(FileSeparators) + 1);
var label = int.Parse(fileName[0].ToString());
var values = new DenseVector(bitmap.Height * bitmap.Width);
for (var i = 0; i < bitmap.Height; i++)
{
for (var j = 0; j < bitmap.Width; j++)
{
{
var color = bitmap.GetPixel(j, i);
var gray = (int)(color.R * 0.3 + color.G * 0.59 + color.B * 0.11);
values[j + i * bitmap.Width] = 1 - gray / 255.0;
}
}
}
if (normalize)
{
var max = values.Maximum();
max /= 2;
values.MapInplace(v => v - max);
}
//values.CoerceZero(0.005);
//values.MapInplace(v => v > 0.995 ? 1 : v);
Vector solution;
if (trainAsEncoder)
{
solution = values;
}
else
{
solution = new DenseVector(10)
{
[label] = 1.0
};
}
var image = new MnistModel
{
Width = bitmap.Width,
Height = bitmap.Height,
Values = values,
FileName = fileName,
Label = label,
ExpectedSolution = solution
};
return(image);
}
public static DenseMatrix ToClassMatrix(this DenseVector x, int?length = null)
{
int len = length.HasValue ? length.Value : (int)x.Maximum() + 1;
var res = DenseMatrix.Create(x.Count, len, (_, __) => 0);
for (int i = 0; i < x.Count; i++)
{
res[i, (int)x[i]] = 1.0;
}
return(res);
}
public SelectorResult Compute(float Percent, int maxrank)
{
if (Percent > 100 || Percent < 0)
{
throw new ArgumentException("");
}
int countA = (int)(Percent * this._x.ColumnCount / 100);
var GroupAIndex = new List <int>();
var GroupBIndex = new List <int>();
if (this._x == null)
{
return(null);
}
var resid = this._x * this._x.Transpose();
var j = this._x.ColumnCount;
var p2 = Math.Round(j * Percent / 100);
int k = 0;
this._x = (Matrix)this._x.Transpose();
var i = maxrank;
var p = new DenseMatrix(i, this._x.ColumnCount);
var told = new DenseMatrix(j, i);
var t = new DenseMatrix(j, i, 1);
var w = new DenseMatrix(i, this._x.ColumnCount);
var l = new DenseVector(i);
for (int h = 0; h < i; h++)
{
double sumt = 0;
for (int ddd = 0; ddd < j; ddd++)
{
sumt += Math.Abs(told[ddd, h] - t[ddd, h]);
}
while ((sumt > 1E-15 && k < 100))
{
told.SetColumn(h, t.Column(h));
k++;
var wt = t.Column(h) * this._x / t.Column(h).DotProduct(t.Column(h));//更新更改
wt = wt / Math.Sqrt(wt.DotProduct(wt));
w.SetRow(h, wt);
t.SetColumn(h, this._x * w.Row(h) / w.Row(h).DotProduct(w.Row(h)));
}
l[h] = k;
k = 0;
var pt = t.Column(h) * this._x / t.Column(h).DotProduct(t.Column(h));
pt = pt / Math.Sqrt(pt.DotProduct(pt));
var tempt = Math.Sqrt(pt.DotProduct(pt));
p.SetRow(h, pt);
t.SetColumn(h, t.Column(h) / tempt);
w.SetRow(h, w.Row(h) / tempt);
this._x = (Matrix)(this._x - t.Column(h).ToColumnMatrix() * p.Row(h).ToRowMatrix());
}
var G = t.Transpose();
var dist = new EuclidianDistance();
var MM = Matlab.PDist((Matrix)G, dist, VectorType.Column);
var M = Matlab.Triu(MM);
double[] maxVal;
int[] maxIdx;
Matlab.Max(M, out maxVal, out maxIdx);
var ttt = new DenseVector(maxVal);
int idx = ttt.MaximumIndex();
double temp = ttt.Maximum();
GroupAIndex.AddRange(new int[] { idx, maxIdx[idx] });
for (int tr = 2; tr < p2; tr++)
{
var tm = new DenseMatrix(GroupAIndex.Count, MM.ColumnCount);
for (int kkk = 0; kkk < GroupAIndex.Count; kkk++)
{
tm.SetRow(kkk, MM.Column(GroupAIndex[kkk]));
foreach (var r in GroupAIndex)
{
tm[kkk, r] = double.NaN;
}
}
int[] midx;
double[] minval;
Matlab.Min(tm, out minval, out midx);
var minttt = new DenseVector(minval);
temp = minttt.Maximum();
idx = minttt.MaximumIndex();
GroupAIndex.Add(idx);
}
GroupAIndex.Sort();
for (int r = 0; r < j; r++)
{
if (!GroupAIndex.Contains(r))
{
GroupBIndex.Add(r);
}
}
return(new SelectorResult()
{
GroupAIndex = GroupAIndex.ToArray(), GroupBIndex = GroupBIndex.ToArray()
});
}