MNIST Neural Network¶
In [2]:
import mnist_loader
training_data, validation_data, test_data = mnist_loader.load_data_wrapper()
In [3]:
len(training_data)
Out[3]:
50000
In [4]:
len(test_data)
Out[4]:
10000
In [5]:
len(validation_data)
Out[5]:
10000
In [6]:
import network
In [7]:
net = network.Network([784,30,10])
In [8]:
net.SGD(training_data, 30, 10, 3.0, test_data=test_data)
Epoch 0: 7291 / 10000, took 2.79 seconds Epoch 1: 7378 / 10000, took 2.75 seconds Epoch 2: 9240 / 10000, took 2.61 seconds Epoch 3: 9313 / 10000, took 2.56 seconds Epoch 4: 9375 / 10000, took 2.55 seconds Epoch 5: 9380 / 10000, took 2.49 seconds Epoch 6: 9362 / 10000, took 2.56 seconds Epoch 7: 9370 / 10000, took 2.62 seconds Epoch 8: 9420 / 10000, took 2.55 seconds Epoch 9: 9447 / 10000, took 2.63 seconds Epoch 10: 9422 / 10000, took 2.65 seconds Epoch 11: 9412 / 10000, took 2.48 seconds Epoch 12: 9459 / 10000, took 2.54 seconds Epoch 13: 9460 / 10000, took 2.68 seconds Epoch 14: 9460 / 10000, took 2.52 seconds Epoch 15: 9468 / 10000, took 2.49 seconds Epoch 16: 9444 / 10000, took 2.64 seconds Epoch 17: 9467 / 10000, took 2.68 seconds Epoch 18: 9478 / 10000, took 2.64 seconds Epoch 19: 9482 / 10000, took 2.57 seconds Epoch 20: 9482 / 10000, took 2.58 seconds Epoch 21: 9481 / 10000, took 2.50 seconds Epoch 22: 9489 / 10000, took 2.53 seconds Epoch 23: 9499 / 10000, took 2.46 seconds Epoch 24: 9496 / 10000, took 2.49 seconds Epoch 25: 9491 / 10000, took 2.61 seconds Epoch 26: 9492 / 10000, took 2.64 seconds Epoch 27: 9488 / 10000, took 2.55 seconds Epoch 28: 9484 / 10000, took 2.47 seconds Epoch 29: 9483 / 10000, took 2.48 seconds
Now we change the number of hidden neurons to 100.
In [9]:
net = network.Network([784, 100, 10])
In [10]:
net.SGD(training_data, 30, 10, 3.0, test_data=test_data)
Epoch 0: 5543 / 10000, took 8.93 seconds Epoch 1: 5644 / 10000, took 8.60 seconds Epoch 2: 5699 / 10000, took 8.46 seconds Epoch 3: 5781 / 10000, took 8.43 seconds Epoch 4: 6572 / 10000, took 7.99 seconds Epoch 5: 6568 / 10000, took 8.07 seconds Epoch 6: 6627 / 10000, took 7.96 seconds Epoch 7: 6654 / 10000, took 7.92 seconds Epoch 8: 6664 / 10000, took 7.93 seconds Epoch 9: 6703 / 10000, took 8.13 seconds Epoch 10: 6673 / 10000, took 7.91 seconds Epoch 11: 6693 / 10000, took 8.24 seconds Epoch 12: 6712 / 10000, took 8.01 seconds Epoch 13: 6759 / 10000, took 8.10 seconds Epoch 14: 6835 / 10000, took 7.92 seconds Epoch 15: 7720 / 10000, took 7.92 seconds Epoch 16: 8647 / 10000, took 8.04 seconds Epoch 17: 8684 / 10000, took 8.04 seconds Epoch 18: 8717 / 10000, took 7.94 seconds Epoch 19: 8728 / 10000, took 8.14 seconds Epoch 20: 8749 / 10000, took 7.98 seconds Epoch 21: 8729 / 10000, took 7.86 seconds Epoch 22: 8721 / 10000, took 7.90 seconds Epoch 23: 8743 / 10000, took 8.21 seconds Epoch 24: 8732 / 10000, took 8.04 seconds Epoch 25: 8743 / 10000, took 8.01 seconds Epoch 26: 8729 / 10000, took 7.95 seconds Epoch 27: 8744 / 10000, took 7.92 seconds Epoch 28: 8754 / 10000, took 8.10 seconds Epoch 29: 8758 / 10000, took 8.11 seconds
In [ ]: