error rate not decrease

[leoxiaobin]

hi, I have used your NIN configure for cifar10 to train the model, but I find that the error rate will not decrease when I am training, do you have any advice when train the NIN model. The error rate is just like this:
9.1... logprob: 2.302583, 0.900000 (35.117 sec)
9.2... logprob: 2.302583, 0.900000 (34.752 sec)
9.3... logprob: 2.302583, 0.900000 (34.942 sec)
9.4... logprob: 2.302583, 0.900000 (34.813 sec)
9.5... logprob: 2.302583, 0.900000 (34.813 sec)
10.1... logprob: 2.302583, 0.900000 (34.811 sec)
10.2... logprob: 2.302583, 0.900000 (34.812 sec)
10.3... logprob: 2.302583, 0.900000 (34.824 sec)
10.4... logprob: 2.302583, 0.900000 (34.822 sec)
10.5... logprob: 2.302583, 0.900000 (34.801 sec)
11.1... logprob: 2.302583, 0.900000 (34.809 sec)
11.2... logprob: 2.302583, 0.900000 (34.828 sec)
11.3... logprob: 2.302583, 0.900000 (34.833 sec)
11.4... logprob: 2.302583, 0.900000 (34.875 sec)

[Lina1991]

Hello, i run the convnet.py and countered the "segmentation fault(core dumped)" at " importing the C++ libmodel " line, i confused why?