blob: 8a9206fafbb992446911beefdac01ded1163f721 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
|
enabling BFGS based optimization **without** curvature calculation
creating cache_file = train-sets/frank.dat.cache
Reading from train-sets/frank.dat
num sources = 1
Num weight bits = 20
learning rate = 10
initial_t = 1
power_t = 0.5
decay_learning_rate = 1
using l2 regularization
m = 7
Allocated 72M for weights and mem
## avg. loss der. mag. d. m. cond. wolfe1 wolfe2 mix fraction curvature dir. magnitude step size time
1 3.313292e+12 9.499654e+14 1.905397e+13 6.440241e+13 2.698544e+12 2.958580e-01 0.690
3 4.946559e+11 1.587411e+13 2.431760e+11 0.500000 0.000000 4.134247e+08 1.000000e+00 1.152
4 4.262257e+11 1.360521e+13 2.087762e+11 0.963277 0.926554 9.978023e+10 1.000000e+00 1.711
5 5.509095e+09 1.685125e+10 8.165054e+08 0.502727 0.005672 4.928468e+09 1.000000e+00 2.350
6 2.722915e+09 9.129735e+09 4.115941e+08 0.848424 0.697668 2.678213e+10 1.000000e+00 3.318
7 5.137694e+06 6.706857e+06 1.018132e+05 0.499525 -0.000686 2.755246e+05 1.000000e+00 4.523
8 5.995715e+06 3.517040e+07 5.294065e+05 -0.644854 -2.289846 (revise x 0.5) 5.000000e-01 4.767
9 5.019540e+06 2.805249e+06 4.237483e+04 0.177600 -0.644881 1.362459e+04 1.000000e+00 6.061
10 4.934764e+06 6.562551e+00 1.255939e-01 0.499927 -0.000122 7.431789e-03 1.000000e+00 7.549
finished run
number of examples = 100000
weighted example sum = 1e+05
weighted label sum = -1.577e+11
average loss = 7.556e+11
best constant = -1.577e+06
total feature number = 400000
|
