数字识别神经网络转换为定点并生成C代码

%加载数据tr = csvread (“train.csv”1 0);%读train.csv子= csvread (“test.csv”1 0);%读test.csv%准备数据n =大小(tr, 1);%的样本数据集目标= tr (: 1);| | % 1列标签目标(目标= = 0)= 10;%使用‘10’现在' 0 'targetsd = dummyvar(目标);%的标签转换成一个虚拟变量输入= tr(:, 2:结束);%的列是预测输入=输入';%的转置输入目标=目标”;%的转置目标targetsd = targetsd ';% '哑变量rng (1);%的再现性c = cvpartition (n,“坚持”、n / 3);%坚持1/3的数据集Xtrain =输入(:,培训(c));% 2/3的输入进行训练Ytrain = targetsd(:,培训(c));% 2/3的培训目标Xtest =输入(:,测试(c));%的1/3的输入进行测试欧美=目标(测试(c));%的1/3的目标检测Ytestd = targetsd(:,测试(c));% 1/3的哑变量进行测试%的列车网络的hiddenLayerSize = 100;网= patternnet (hiddenLayerSize);(净,tr) =火车(网,Xtrain, Ytrain);视图(净);输出=净(Xtest);错误= gsubtract(欧美、输出);性能=执行(欧美,净输出);图中,plotperform (tr);

nnet.guis.closeAllViews ();

sys_name = gensim(网络,“名字”,“mTrainedNN”);

模型=“fxpdemo_mnist_classification”;system_under_design =[模型' /模式识别神经网络的];baseline_output =[模型' / yarr '];open_system(模型);

转换器= DataTypeWorkflow.Converter (system_under_design);

converter.applySettingsFromShortcut (收集使用双覆盖范围的);collect_ranges = converter.CurrentRunName;sim_out = converter.simulateSystem ();

plotConfusionMatrix (sim_out baseline_output system_under_design,“之前分类率转换”);

ps = DataTypeWorkflow.ProposalSettings;转换器。proposeDataTypes (collect_ranges、ps);

converter.applyDataTypes (collect_ranges);

converter.applySettingsFromShortcut (的范围与指定的数据类型集合);sim_out = converter.simulateSystem ();

plotConfusionMatrix (sim_out baseline_output system_under_design,“定点后分类率转换”);

block_path = [system_under_design“图层1 / tansig”];p = FunctionApproximation.Problem (block_path);p.Options。字= 16;p.Options。BreakpointSpecification =“EvenPow2Spacing”;解决方案= p.solve;solution.replaceWithApproximate;

block_path = [system_under_design/ 2层/ softmax /经验的];p = FunctionApproximation.Problem (block_path);p.Options。字= 16;p.Options。BreakpointSpecification =“EvenPow2Spacing”;

p。InputLowerBounds = -40;p。InputUpperBounds = 0;解决方案= p.solve;solution.replaceWithApproximate;

converter.applySettingsFromShortcut (converter.ShortcutsForSelectedSystem {2});sim_out = converter.simulateSystem;plotConfusionMatrix (sim_out baseline_output system_under_design,分类率函数替换后的);