NNUE Poll

[Rebel]

A missed poll question - I'd like to adopt NNUE but I lack the computer power it needs.

[Chessnut1071]

NNUE is very effective at optimizing the training set; however, the biggest issue with NNUE is over fitting, since a larger number of nodes is usually required to be optimized compared with standard evaluation. NNUE is often less effective on games not in its training set. There's a simple example of AI preforming poorly compared to a statistical technique called discriminant analysis. A $200,000 NNUE software program classified all but 6 classifications of Fisher's (1936) well-known Iris problem correctly after 12 minutes (720 seconds) run time. Standard discriminant analysis found all bur 2 in .001 seconds.

[dangi12012]

Standard discriminant analysis found all bur 2 in .001 seconds.

Id like to see the source on that.

[Chessnut1071]

Standard discriminant analysis found all bur 2 in .001 seconds.

Id like to see the source on that.

Any standard statistical package with a Proc Discrim will run the data below. The test was between an AI program by the US Dept of The Navy at IBM HQ in Texas. The demo was an attempt to show the Navy how they were misusing analytical tools. Use any AI program you like, I have not seen one that produced more correct classifications than standard Discriminant Analysis. Most are notorious for over-fitting the training set. BTW, if you're optimizing evaluation functions for chess, I developed a nonlinear program using a modified Hooke & Jeeves algorithm that is slower, but more accurate and precise than gradient methods. Enjoy

Fisher Iris data
No. SepalLen SepalWid PetalLen PetalWid Species
1 50 33 14 2 1achieve better classification
2 67 31 56 24 3
3 69 31 51 23 3
4 46 36 10 2 1
5 65 30 52 20 3
6 58 27 51 19 3
7 57 28 45 13 2
8 63 33 47 16 2
9 49 25 45 17 3
10 70 32 47 14 2
11 48 31 16 2 1
12 63 25 50 19 3
13 49 36 14 1 1
14 44 32 13 2 1
15 58 26 40 12 2
16 63 27 49 18 3
17 50 23 33 10 2
18 51 38 16 2 1
19 64 28 56 22 3
20 63 28 51 15 3
21 62 22 45 15 2
22 61 30 46 14 2
23 56 25 39 11 2
24 68 32 59 23 3
25 62 34 54 23 3
26 67 33 57 25 3
27 55 35 13 2 1
28 64 32 45 15 2
29 59 30 51 18 3
30 64 32 53 23 3
31 54 30 45 15 2
32 67 33 57 21 3
33 44 30 13 2 1
34 47 32 16 2 1
35 72 32 60 18 3
36 61 30 49 18 3
37 65 28 46 15 2
38 46 34 14 3 1
39 59 32 48 18 2
40 60 27 51 16 2
41 65 30 55 18 3
42 51 33 17 5 1
43 77 38 67 22 3
44 76 30 66 21 3
45 67 30 52 23 3
46 61 28 40 13 2
47 55 24 38 11 2
48 52 34 14 2 1
49 79 38 64 20 3
50 50 35 16 6 1
51 77 28 67 20 3
52 55 26 44 12 2
53 48 30 14 3 1
54 48 34 19 2 1
55 50 30 16 2 1
56 64 28 56 21 3
57 51 38 19 4 1
58 49 30 14 2 1
59 58 27 41 10 2
60 60 29 45 15 2
61 50 36 14 2 1
62 58 27 51 19 3
63 54 34 15 4 1
64 64 31 55 18 3
65 49 24 33 10 2
66 55 42 14 2 1
67 60 22 50 15 3
68 52 27 39 14 2
69 44 29 14 2 1
70 58 27 39 12 2
71 69 32 57 23 3
72 59 30 42 15 2
73 56 28 49 20 3
74 67 25 58 18 3
75 63 23 44 13 2
76 63 25 49 15 2
77 51 25 30 11 2
78 69 31 54 21 3
79 72 36 61 25 3
80 56 29 36 13 2
81 68 30 55 21 3
82 48 30 14 1 1
83 57 38 17 3 1
84 66 30 44 14 2
85 51 37 15 4 1
86 67 30 50 17 2
87 50 32 12 2 1
88 43 30 11 1 1
89 67 31 44 14 2
90 51 35 14 2 1
91 50 34 16 4 1
92 57 26 35 10 2
93 77 30 61 23 3
94 57 29 42 13 2
95 52 41 15 1 1
96 60 30 48 18 3
97 56 27 42 13 2
98 49 31 15 2 1
99 54 39 17 4 1
100 60 34 45 16 2
101 50 20 35 10 2
102 47 32 13 2 1
103 62 29 43 13 2
104 51 34 15 2 1
105 60 22 40 10 2
106 49 31 15 1 1
107 54 37 15 2 1
108 61 28 47 12 2
109 57 28 41 13 2
110 54 39 13 4 1
111 65 32 51 20 3
112 69 31 49 15 2
113 55 25 40 13 2
114 45 23 13 3 1
115 51 38 15 3 1
116 68 28 48 14 2
117 52 35 15 2 1
118 63 33 60 25 3
119 61 26 56 14 3
120 58 40 12 2 1
121 62 28 48 18 3
122 56 30 45 15 2
123 46 32 14 2 1
124 57 44 15 4 1
125 63 34 56 24 3
126 72 30 58 16 3
127 71 30 59 21 3
128 63 29 56 18 3
129 57 30 42 12 2
130 77 26 69 23 3
131 66 29 46 13 2
132 50 34 15 2 1
133 55 24 37 10 2
134 46 31 15 2 1
135 74 28 61 19 3
136 50 35 13 3 1
137 73 29 63 18 3
138 67 31 47 15 2
139 56 30 41 13 2
140 64 29 43 13 2
141 65 30 58 22 3
142 51 35 14 3 1
143 61 29 47 14 2
144 64 27 53 19 3
145 48 34 16 2 1
146 57 25 50 20 3
147 55 23 40 13 2
148 54 34 17 2 1
149 58 28 51 24 3
150 53 37 15 2 1