In [1]:
import caffe
caffe.set_mode_cpu()
import numpy as np
import matplotlib.pyplot as plt
plt.rcParams['image.cmap'] = 'gray'
%matplotlib inline
In Caffe models get specified in separate protobuf files.
Additionally a solver has to be specified, that determines training parameters.
Instantiate the solver and train the network.
In [3]:
solver = caffe.SGDSolver('mnist_solver.prototxt')
solver.net.forward()
Out[3]:
In [8]:
niter = 2500
test_interval = 100
# losses will also be stored in the log
train_loss = np.zeros(niter)
test_acc = np.zeros(int(np.ceil(niter / test_interval)))
output = np.zeros((niter, 8, 10))
# the main solver loop
for it in range(niter):
solver.step(1) # SGD by Caffe
# store the train loss
train_loss[it] = solver.net.blobs['loss'].data
# store the output on the first test batch
# (start the forward pass at conv1 to avoid loading new data)
solver.test_nets[0].forward(start='conv2d_1')
output[it] = solver.test_nets[0].blobs['dense_2'].data[:8]
# run a full test every so often
# (Caffe can also do this for us and write to a log, but we show here
# how to do it directly in Python, where more complicated things are easier.)
if it % test_interval == 0:
print ('Iteration', it, 'testing...')
correct = 0
test_iter = 100
for test_it in range(test_iter):
solver.test_nets[0].forward()
correct += sum(solver.test_nets[0].blobs['dense_2'].data.argmax(1)
== solver.test_nets[0].blobs['label'].data)
test_acc[it // test_interval] = correct / (64 * test_iter)
In [9]:
_, ax1 = plt.subplots()
ax2 = ax1.twinx()
ax1.plot(np.arange(niter), train_loss)
ax2.plot(test_interval * np.arange(len(test_acc)), test_acc, 'r')
ax1.set_xlabel('iteration')
ax1.set_ylabel('train loss')
ax2.set_ylabel('test accuracy')
ax2.set_title('Test Accuracy: {:.2f}'.format(test_acc[-1]))
Out[9]:
The weights are saved in a .caffemodel file.
In [11]:
solver.net.save('mnist.caffemodel')