This document discusses using a convolutional neural network (CNN) to classify handwritten digits from the MNIST dataset. It describes training a CNN model on MNIST data for 3 epochs, then using the trained model to predict the digit in an input image. The CNN architecture includes convolution and max pooling layers followed by a fully-connected layer. The model is trained on GPU and saved, then used to predict digits in new images.

1. Learning
Hands-on
Deep

10. curl https://github.com/pfnet/chainer/archive/v1.19.0.tar.gz -o v1.19.tar.gz
tar xzf v1.19.0.tar.gz
cd chainer-1.19.0/examples/mnist
python train_mnist.py -e 5 -u 200
GPU: -1
# unit: 200
# Minibatch-size: 100
# epoch: 5
epoch main/loss validation/main/loss main/accuracy validation/main/accur
1 0.274201 0.129138 0.921567 0.9604
2 0.106627 0.0927404 0.968 0.9691
3 0.0719209 0.0824399 0.9775 0.9748
4 0.0537356 0.0790009 0.983133 0.9758
5 0.0394959 0.0797915 0.987283 0.9753
↑ ↑↑ ↑ ↑

11. class MLP(chainer.Chain):
def __init__(self, n_units, n_out):
super(MLP, self).__init__(
# the size of the inputs to each layer will be inferred
l1=L.Linear(None, n_units), # n_in -> n_units
l2=L.Linear(None, n_units), # n_units -> n_units
l3=L.Linear(None, n_out), # n_units -> n_out
)
def __call__(self, x):
h1 = F.relu(self.l1(x))
h2 = F.relu(self.l2(h1))
return self.l3(h2)
13

12. def main():
parser = argparse.ArgumentParser(description='Chainer example: MNIST')
parser.add_argument('--batchsize', '-b', type=int, default=100,
help='Number of images in each mini-batch')
parser.add_argument('--epoch', '-e', type=int, default=20,
help='Number of sweeps over the dataset to train')
parser.add_argument('--gpu', '-g', type=int, default=-1,
help='GPU ID (negative value indicates CPU)')
parser.add_argument('--out', '-o', default='result',
help='Directory to output the result')
parser.add_argument('--resume', '-r', default='',
help='Resume the training from snapshot')
parser.add_argument('--unit', '-u', type=int, default=1000,
help='Number of units')
args = parser.parse_args()
python train_mnist.py -e 5 -u 200
29

13. optimizer = chainer.optimizers.Adam()
60

16.
# Run the training
trainer.run()
# save the tranined model
chainer.serializers.save_npz('linear.model', model)
if __name__ == '__main__':
main()
103

17. import numpy as np
# Network definition
class MLP(chainer.Chain):
… train_mnist.py …
def main():
parser = argparse.ArgumentParser(description='Chainer example: MNIST')
parser.add_argument('--unit', '-u', type=int, default=1000, help='Number of units')
parser.add_argument('--number', '-n', type=int, default=1, help='mnist index')
args = parser.parse_args()
train, test = chainer.datasets.get_mnist()
index = min(args.number,9999)
targetNumber = test[index][0].reshape(-1,784)
targetAnswer = test[index][1]
model = L.Classifier(MLP(args.unit, 10))
chainer.serializers.load_npz('linear.model', model)
# Results
x = chainer.Variable(targetNumber)
v = model.predictor(x)
print("mnistIndex:",args.number,"answer:", targetAnswer ,"predict:", np.argmax(v.data))
main()
13

18. python predict_mnist.py -u 200 -n 7777
↑ ↑
9999
mnistIndex: 7777 answer: 5 predict: 5

20. import numpy as np
from PIL import Image
# Network definition
class MLP(chainer.Chain):
… train_mnist.py …
def main():
parser = argparse.ArgumentParser(description='Chainer example: MNIST')
parser.add_argument('--unit', '-u', type=int, default=1000, help='Number of units')
parser.add_argument('--name', '-n', type=str, default="1.png", help='file name')
args = parser.parse_args()
myNumber = Image.open(args.name).convert("L")
myNumber = 1.0 - np.asarray(myNumber, dtype="float32") / 255
myNumber = myNumber.reshape((1, 784))
model = L.Classifier(MLP(args.unit, 10))
chainer.serializers.load_npz('linear.model', model)
# Results
x = chainer.Variable(myNumber)
v = model.predictor(x)
print("fileName:", args.name, "predict:", np.argmax(v.data))
main()
13

21. python predict_handwritten.py -u 200 -n 2.jpg
↑ ↑
fileName: 2.jpg predict: 2

23. class CNN(chainer.Chain):
def __init__(self, train=True):
super(CNN, self).__init__(
conv1=L.Convolution2D(1, 32, 5),
conv2=L.Convolution2D(32, 64, 5),
l1=L.Linear(1024, 10),
)
self.train = train
def __call__(self, x):
h = F.max_pooling_2d(F.relu(self.conv1(x)), 2)
h = F.max_pooling_2d(F.relu(self.conv2(h)), 2)
return self.l1(h)
13
MLP —> CNN

24. 43
# Set up a neural network to train
model = L.Classifier(CNN())
if args.gpu >= 0:
chainer.cuda.get_device(args.gpu).use() # Make a specified GPU current
model.to_gpu() # Copy the model to the GPU
# Setup an optimizer
optimizer = chainer.optimizers.Adam()
optimizer.setup(model)
# Load the MNIST dataset
train, test = chainer.datasets.get_mnist(ndim=3)
<— CNN
↑ MNIST CNN [ ]
python train_mnist_CNN.py -e 3

25. # Network definition
class CNN(chainer.Chain):
… Train_mnist_CNN.py …
def main():
parser = argparse.ArgumentParser(description='Chainer example: MNIST')
parser.add_argument('--unit', '-u', type=int, default=1000, help='Number of units')
parser.add_argument('--name', '-n', type=str, default="1.png", help='file name')
args = parser.parse_args()
model = L.Classifier(CNN())
myNumber = Image.open(args.name).convert("L")
myNumber = 1.0 - np.asarray(myNumber, dtype="float32") / 255
myNumber = myNumber.reshape((1,1,28,28))
chainer.serializers.load_npz('cnn.model', model)
# Results
x = chainer.Variable(myNumber)
v = model.predictor(x)
print(args.name , np.argmax(v.data))
main()
13
python predict_handwritten.py -n 2.jpg