sureli / neurops Goto Github PK

While growing LeNet for just second iteration, this error terminates the program.

I think earlier version did not produce this error for LeNet.

So majorly VGG-11 is showing 0 in front of "neurons to add". Could it be please confirmed whether grow algorithm running is fine or not in this because such an issue was not observed for LeNet.

Just to add, these are few results of last 2 iterations:

Validation: Average loss: 0.0144, Accuracy: 3467/5000 (69.34%)
Layer 0 score: 21/25, neurons to add: 0
Layer 1 score: 34/36, neurons to add: 0
Layer 2 score: 63/67, neurons to add: 0
Layer 3 score: 61/65, neurons to add: 0
Layer 4 score: 124/133, neurons to add: 0
Layer 5 score: 112/128, neurons to add: 0
Layer 6 score: 105/128, neurons to add: 0
Layer 7 score: 78/128, neurons to add: 0
Layer 8 score: 304/1024, neurons to add: 0
Layer 9 score: 475/1024, neurons to add: 0
The grown model now has 1787097 effective parameters.
Validation after growing: Average loss: 0.0143, Accuracy: 3467/5000 (69.34%)
Train Epoch: 0 [0/45000 (0%)] Loss: 0.047333
Train Epoch: 0 [12800/45000 (28%)] Loss: 0.073766
Train Epoch: 0 [25600/45000 (57%)] Loss: 0.026595
Train Epoch: 0 [38400/45000 (85%)] Loss: 0.014527