Matlab实现手写体识别程序，优秀源代码！

function train(hidden_size, lr) % 完整训练过程实现 % % :param hidden_size: 隐藏层单元个数 % :param lr: 学习率 % 程序实现1：参数设置（可以尝试自己修改）===================================== input_size = 784; output_size = 10; train_num = 60000; epoch = 5; batch_size = 100; log_freq = 100; steps_per_epoch = train_num / batch_size; % 程序结束1================================================================= % 读取mnist数据集 [test_x, test_y, train_x, train_y] = read_mnist(); % 初始化权重值 W1 = normrnd(0, 1e-2, [input_size, hidden_size]); b1 = zeros(1, hidden_size); W2 = normrnd(0, 1e-2, [hidden_size, output_size]); b2 = zeros(1, output_size); % log loss_list = zeros(epoch * steps_per_epoch, 1); err_list = zeros(epoch * steps_per_epoch, 1); times = 1; % 迭代训练 for i=1:epoch % 程序实现2：随机小批量训练集读取======================================== % 提示：可以了解一下randperm函数 for j=1:steps_per_epoch train_data = train_label = % 程序结束2============================================================== % 单步训练，权重更新 [loss, score, W1, b1, W2, b2] = train_step(train_data, train_label, W1, b1, W2, b2, lr); % log % 程序实现3：计算错误率============================================== err = % 程序结束3========================================================= loss_list(times) = loss; err_list(times) = err; if mod(times, log_freq) == 0 fprintf('epoch: %2d | steps: %6d | loss: %6.4f | error: %.4f\n', i, times, loss, err); end times = times + 1; end end % 输出训练结果 save weights.mat W1 b1 W2 b2 log_steps = 1:epoch * steps_per_epoch; figure(); plot(log_steps, loss_list) title('loss curve') xlabel('steps') ylabel('loss') figure(); plot(log_steps, err_list); title('error curve') xlabel('steps') ylabel('error') % 请记录下你的隐藏层个数和学习率============================================= % hidden size = % learning rate = %==========================================================================