feiyuhug / lenet-5 Goto Github PK

Hi,

I cannot fine the softmax_outputlayer.py. Can you upload it at your convinence？

Thank you very much.

Good afternoon, dear respective feiyuhug!

Could you please say why I have got 0% of recognized images on 500 Testing MNIST subset after training. I have launched your LeNet-5 with following parameters in covnet.py:
self.covlay1.calc_maps(mapset)
self.poolinglay2.calc_maps(self.covlay1.maps)
self.covlay3.calc_maps(self.poolinglay2.maps, True)
self.poolinglay4.calc_maps(self.covlay3.maps)
self.covlay5.calc_maps(self.poolinglay4.maps)
self.fclay6.calc_maps(self.covlay5.maps)
#self.outputlay7.softmax(self.fclay6.maps)
self.outputlay7.rbf(self.fclay6.maps, mapclass)

There is no any Softmax.py file in your build so I have set the last output layer as classic RBF. Then in net_train.py I launched your project in a following way:

train_covnet = CovNet()
train_net(train_covnet, logfile, 2, [0.005, 0.001], 60000)
test_net(train_covnet, logfile, 500)

Training goes fine with out errors, and once the Method test_net(train_covnet, logfile, 500) launch after passing on 500 testing samples I receive 0% percentage of recognized images, all 500 testing images after training was recognized wrong. Could you please say this occurs because you have not yet finished your project? Could you please say where can I find a final version of your project or can you say why softmax layer is absent in your project, maybe because of this I get 0% of recognized images? Thank you in advance