wrong result!

[andyyuan78]

running 24 hours, the result are always valid error 100.000%, best valid error 100.000%

ubgpu@ubgpu:~/github/trainingRNNs$ sudo python RNN.py [sudo] password for ubgpu: Using gpu device 0: GeForce GTX 970 /usr/local/lib/python2.7/dist-packages/theano/scan_module/scan_perform_ext.py:133: RuntimeWarning: numpy.ndarray size changed, may indicate binary incompatibility from scan_perform.scan_perform import * Starting to train Iter 0000020 : train nnl 1.760, valid error 100.000%, best valid error 100.000%, average gradient norm 1.672, rho_Whh 1.06, Omega 0.02, alpha 2.000, steps in the past 1.050 Iter 0000040 : train nnl 1.285, valid error 100.000%, best valid error 100.000%, average gradient norm 2.910, rho_Whh 1.22, Omega 0.17, alpha 2.000, steps in the past 1.000 Iter 0000060 : train nnl 0.897, valid error 100.000%, best valid error 100.000%, average gradient norm 2.465, rho_Whh 1.34, Omega 0.38, alpha 2.000, steps in the past 1.000 Iter 0000080 : train nnl 0.584, valid error 100.000%, best valid error 100.000%, average gradient norm 1.567, rho_Whh 1.44, Omega 0.54, alpha 2.000, steps in the past 1.000 Iter 0000100 : train nnl 0.484, valid error 100.000%, best valid error 100.000%, average gradient norm 0.788, rho_Whh 1.50, Omega 0.65, alpha 2.000, steps in the past 1.000 Iter 0000120 : train nnl 0.429, valid error 100.000%, best valid error 100.000%, average gradient norm 0.564, rho_Whh 1.52, Omega 0.69, alpha 2.000, steps in the past 1.000 Iter 0000140 : train nnl 0.395, valid error 100.000%, best valid error 100.000%, average gradient norm 0.464, rho_Whh 1.53, Omega 0.71, alpha 2.000, steps in the past 1.000 ....................

Iter 2131680 : train nnl 0.150, valid error 100.000%, best valid error 099.990%, average gradient norm 0.036, rho_Whh 1.84, Omega 0.02, alpha 2.000, steps in the past 1.000 Iter 2131700 : train nnl 0.138, valid error 100.000%, best valid error 099.990%, average gradient norm 0.034, rho_Whh 1.84, Omega 0.02, alpha 2.000, steps in the past 1.000 Iter 2131720 : train nnl 0.127, valid error 100.000%, best valid error 099.990%, average gradient norm 0.033, rho_Whh 1.84, Omega 0.02, alpha 2.000, steps in the past 1.000 Iter 2131740 : train nnl 0.111, valid error 100.000%, best valid error 099.990%, average gradient norm 0.027, rho_Whh 1.84, Omega 0.01, alpha 2.000, steps in the past 1.000 Iter 2131760 : train nnl 0.115, valid error 100.000%, best valid error 099.990%, average gradient norm 0.029, rho_Whh 1.84, Omega 0.01, alpha 2.000, steps in the past 1.000 Iter 2131780 : train nnl 0.128, valid error 100.000%, best valid error 099.990%, average gradient norm 0.031, rho_Whh 1.84, Omega 0.01, alpha 2.000, steps in the past 1.000

---force stop after 24 hours.