用pycaffe绘制训练过程的loss和accuracy曲线

#!/usr/bin/env python
# 导入绘图库
from pylab import *
import matplotlib.pyplot as plt


# 导入"咖啡"
import caffe


# 设置为gpu模式
caffe.set_device(0)
caffe.set_mode_gpu()


# 使用SGDSolver，即随机梯度下降算法
solver = caffe.SGDSolver('lenet_solver_sgd.prototxt')


# 等价于solver文件中的max_iter，即最大解算次数
niter = 10000
# 每隔100次收集一次数据
display_iter = 100


# 每次测试进行100次解算，根据test用例数量和batch_size得出
test_iter = 100
# 每500次训练进行一次测试（100次解算），根据train用例数量和batch_size得出
test_interval = 500


# train loss
train_loss = zeros(ceil(niter * 1.0 / display_iter))
# test loss
test_loss = zeros(ceil(niter * 1.0 / test_interval))
# test accuracy
test_acc = zeros(ceil(niter * 1.0 / test_interval))


# iteration 0，不计入
solver.step(1)


# 辅助变量
_train_loss = 0; _test_loss = 0; _accuracy = 0
# 进行解算
for it in range(niter):
    # 进行一次解算
    solver.step(1)
    # 计算train loss
    _train_loss += solver.net.blobs['loss'].data
    if it % display_iter == 0:
        # 计算平均train loss
        train_loss[it // display_iter] = _train_loss / display_iter
        _train_loss = 0


    if it % test_interval == 0:
        for test_it in range(test_iter):
            # 进行一次测试
            solver.test_nets[0].forward()
            # 计算test loss
            _test_loss += solver.test_nets[0].blobs['loss'].data
            # 计算test accuracy
            _accuracy += solver.test_nets[0].blobs['accuracy'].data
        # 计算平均test loss
        test_loss[it / test_interval] = _test_loss / test_iter
        # 计算平均test accuracy
        test_acc[it / test_interval] = _accuracy / test_iter
        _test_loss = 0
        _accuracy = 0


# 绘制train loss、test loss和accuracy曲线
print '\nplot the train loss and test accuracy\n'
_, ax1 = plt.subplots()
ax2 = ax1.twinx()


# train loss -> 绿色
ax1.plot(display_iter * arange(len(train_loss)), train_loss, 'g')
# test loss -> 黄色
ax1.plot(test_interval * arange(len(test_loss)), test_loss, 'y')
# test accuracy -> 红色
ax2.plot(test_interval * arange(len(test_acc)), test_acc, 'r')


ax1.set_xlabel('iteration')
ax1.set_ylabel('loss')
ax2.set_ylabel('accuracy')
plt.show()