Gamma Distributioon Fit stalls for some arrays

[KaptenJon]

Gamma Distribution fit function stall for this array.
1275.56
1239.44
1237.92
1237.22
1237.1
1238.41
1238.62
1237.05
1237.19
1236.51
1264.6
1238.19
1237.39
1235.79
1236.53
1236.8
1238.06
1236.5
1235.32
1236.44
1236.58
1236.3
1237.91
1238.6
1238.49
1239.21
1238.57
1244.63
1236.06
1236.4
1237.88
1237.56
1236.66
1236.59
1236.53
1236.32
1238.29
1237.79
1237.86
1236.42
1236.23
1236.37
1237.18
1237.63
1245.8
1238.04
1238.55
1238.39
1236.75
1237.07
1250.78
1238.6
1238.36
1236.58
1236.82
1238.4
1257.68
1237.78
1236.52
1234.9
1237.9
1238.58
1238.12
1237.89
1236.54
1236.55
1238.37
1237.29
1237.64
1236.8
1237.73
1236.71
1238.23
1237.84
1236.26
1237.58
1238.31
1238.4
1237.08
1236.61
1235.92
1236.41
1237.89
1237.98
1246.75
1237.92
1237.1
1237.97
1238.69
1237.05
1236.96
1239.44
1238.49
1237.88
1236.01
1236.57
1236.44
1235.76
1237.62
1238
1263.14
1237.66
1237
1236
1261.96
1238.58
1237.77
1237.06
1236.31
1238.63
1237.23
1236.85
1236.23
1236.46
1236.9
1237.85
1238
1237.02
1236.19
1236.05
1235.73
1258.3
1235.98
1237.76
1246.93
1239.1
1237.72
1237.67
1236.79
1237.61
1238.41
1238.29
1238.11
1237
1236.52
1236.6
1236.31
1237.77
1238.58
1237.88
1247.35
1236.14
1236.83
1236.15
1237.93
1238.16
1237.34
1236.78
1238.66
1237.76
1237.19
1236.7
1236.04
1236.66
1237.86
1238.54
1238.05
1238.41
1236.94
1240.95
1261.01
1237.72
1237.91
1238.2
1235.68
1236.89
1235.12
1271.31
1236.97
1270.76
1238.52
1238.19
1238.6
1237.16
1236.72
1236.71
1237.14
1238.48
1237.95
1237.42
1235.86
1236.39
1236.13
1236.58
1237.95
1237.76
1237.39
1238.16
1236.31
1236.41
1236.12
1238.7
1236.48
1237.84
1236.38
1237.95
1238.48
1236.51
1236.56

[KaptenJon]

Did some tests. It stall in the while loop calculating the k.
It comes into a situation where the values repeats

The problem is that double.Epsilon is the smallest number available. however double has not the same accuracy. I think the code should be

double epsilon = 0;
do
{
oldK = newK;
newK = oldK - (Math.Log(newK) - Accord.Math.Gamma.Digamma(newK) - s) / ((1 / newK) - Accord.Math.Gamma.Trigamma(newK));
epsilon = Math.Max(Math.Abs(newK)* 1E-14, Math.Abs(oldK)) * 1E-14;
}
while (Math.Abs(oldK - newK) > epsilon)

[cesarsouza]

Thank you very much for reporting the issue! I have replaced the convergence check with a more appropriate method that handles more corner cases.

[cesarsouza]

Fixed on release 3.2.