本文主要是介绍Matlab中使用FLOPS计算代码复杂度（浮点数运算次数）方法，希望对大家解决编程问题提供一定的参考价值，需要的开发者们随着小编来一起学习吧！
参考网址
Counting the Floating Point Operations (FLOPS) - File Exchange - MATLAB Central (mathworks.cn)
方法介绍
第1步
准备好需要计算浮点数运算次数的Matlab的.m文件。需要注意的是，尽量让代码结构尽可能简单，只调用常见的函数运算、矩阵操作等。如果代码中必须要调用自己构建的函数，则必须通过子函数将它们内部化，以便后续步骤也可以对它们进行解析。
第2步
在.m文件中代码的最后，将所有变量保存在MAT文件中，只要一行代码即可：
save "MATfileName"如果代码中包含子函数，则上述代码改为：
save('MATfileName','-append')第3步
分析Matlab代码，代码如下(本文最后会讲解官方示例)：
profile on;
运行fileName;
profileStruct=profile('info');第4步
调用FLOPS函数并计算浮点数运算次数
[flopTotal,Details]=FLOPS(fileName,MATfileName,profileStruct);官方示例
1.脚本（Script）
脚本示例：
% This is just a test of FLOPS counting% Matrix dimensions
m = 3;
n = 4;
k = 5;% Matrix generation
A = rand(m,n);
B = ones(m,n);
C = randn(n,k) <= 0.5;% Test plus, minus, multiplication and division
D = A + B;
E = A * C;
F = ((A .* (A + B)) ./ (A-B)) * C;
G = bsxfun(@minus,A,mean(A));% Test linear algebra
P = rand(m);
PP = P * P';
P = chol(P*P');
[L,U] = lu(P);
[Q,R] = qr(P);
P = inv(P);
x = P \ rand(m,1);% Test statistics and math function
for r = 1:mS = sum(A);S = sin(A+2*B);
end% Test user supplied rules
R = mod(randi(100,m,n), randi(100,m,n));
g = gamma(A);% Save all variables in a MAT file for FLOPS counting
save exampleScriptMAT计算示例：
profile on
exampleScript
profileStruct = profile('info');
[flopTotal,Details] = FLOPS('exampleScript','exampleScriptMAT',profileStruct);2.函数（Function，包含有子函数）
脚本示例：
function [Beta_draws,ME1,ME2] = exampleFun(Y,X,ndraws,burn_in)% Purpose:
% Bayesian Estimate of the Probit model and the marginal effects
% -----------------------------------
% Model:
% Yi* = Xi * Beta + ui , where normalized ui ~ N(0,1)
% Yi* is unobservable.
% If Yi* > 0, we observe Yi = 1; If Yi* <= 0, we observe Yi = 0
% -----------------------------------
% Algorithm:
% Gibbs sampler. Proper prior Beta ~ N(mu,V).
% Posterior Beta has conjugate normal.
% Posterior latent variable follows truncated normal.
% -----------------------------------
% Usage:
% Y = dependent variable (n * 1 vector)
% X = regressors (n * k matrix)
% ndraws = number of draws in MCMC
% burn_in = number of burn-in draws in MCMC
% -----------------------------------
% Returns:
% Beta_draws = posterior draws of coefficients corresponding to the k regressors
% ME1 = marginal effects (average data)
% ME2 = marginal effects (individual average)
% -----------------------------------
% Notes:
% Probit model is subject to normalization.
% The variance of disturbances is set to 1, and a constant is added to X.
%
% Version: 06/2012
% Written by Hang Qian, Iowa State University
% Contact me: matlabist@gmail.comif nargin<2; error('Incomplete data.'); end
if nargin<3; ndraws = 300; end
if nargin<4; burn_in = ndraws * 0.5; endMissingValue = any(isnan([Y,X]),2);
if any(MissingValue)disp('There are missing values in your data.')disp(['Discard observations: ',num2str(find(MissingValue'))])FullValue = ~MissingValue; Y = Y(FullValue); X = X(FullValue,:);
end[nobs,nreg] = size(X);%----------------------------------------
% Prior distribution settings
% Beta ~ N(mu,V)
% You may change the hyperparameters here if needed
prior_mu = zeros(nreg,1);
prior_V = 100 * eye(nreg);
%-----------------------------------------Beta_draws = zeros(nreg,ndraws-burn_in);
Z = X * ((X'*X)\(X'*Y));
XX = X' * X;
inv_prior_V = inv(prior_V);
truncate_lower = -999 * (Y == 0);
truncate_upper = 999 * (Y == 1);for r = 1:ndrawsbeta_D = inv(XX + inv_prior_V);beta_d = X' * Z + inv_prior_V * prior_mu; %#ok<MINV>P = chol(beta_D);Beta_use = beta_D * beta_d + P' * randn(nreg,1); %#ok<MINV>Z = TN_RND(X*Beta_use,1,truncate_lower,truncate_upper,nobs);if r > burn_in Beta_draws(:, r - burn_in) = Beta_use;end
endBeta_mean = mean(Beta_draws,2);
Beta_std = std(Beta_draws,0,2);% ME1 = normpdf(mean(X)*Beta_mean,0,1) * Beta_mean;
% ME2 = mean(normpdf(X*Beta_mean,0,1)) * Beta_mean;
ME1 = 1/sqrt(2*pi)*exp(-0.5*(mean(X)*Beta_mean).^2) * Beta_mean;
ME2 = mean(1/sqrt(2*pi)*exp(-0.5*(X*Beta_mean).^2)) * Beta_mean;result = cell(nreg + 1,5);
result(1,:) = {'Coeff.','Post. mean','Post. std','ME(avg. data)','ME(ind. avg.)'};
for m = 1:nregresult(m + 1,1) = {['C(',num2str(m),')']};result(m + 1,2:5) = {Beta_mean(m),Beta_std(m),ME1(m),ME2(m)};
enddisp(' ')
disp(result)save('exampleFunMat','-append')end%-------------------------------------------------------------------------
% Subfunction
function sample = TN_RND(mu,sigma,lb,ub,ndraws)% Purpose:
% Generate random numbers from truncated normal distribution
% TN(lb,ub) (mu, sigma)
% -----------------------------------
% Density:
% f(x) = 1/(Phi(ub)-Phi(lb)) * phi(x,mu,sigma)
% -----------------------------------
% Algorithm:
% Inverse CDF
% -----------------------------------
% Usage:
% mu = location parameter
% sigma = scale parameter
% lb = lower bound of the random number
% ub = upper bound of the random number
% ndraws = number of draws
% -----------------------------------
% Returns:
% sample = random numbers from TN(lb,ub) (mu, sigma)
% -----------------------------------
% Notes:
% 1. If at least one of the arguments mu,sigma,lb,ub are vectors/matrix,
% It will return a vector/matrix random numbers with conformable size.
% 2. If there is no lower/upper bound, use Inf or some large number instead
%
% Version: 06/2012
% Written by Hang Qian, Iowa State University
% Contact me: matlabist@gmail.comif nargin < 4; ub = 999;end
if nargin < 3; lb = -999;end
if nargin < 2; sigma = 1;end
if nargin < 1; mu = 0;endprob_ub = normcdf(ub,mu,sigma);
prob_lb = normcdf(lb,mu,sigma);
prob_diff = prob_ub - prob_lb;ndraws_check = length(prob_diff);
if nargin < 5 | ndraws_check > 1 %#ok<OR2>ndraws = ndraws_check;U = prob_diff;U(:) = rand(ndraws,1);
elseU = rand(ndraws,1);
endU_rescale = prob_lb + U .* prob_diff;
sample = norminv(U_rescale,mu,sigma);save('exampleFunMat')end可以看到，exampleFun函数中TN_RND函数是自己构建的一个函数，且这个函数必须调用，因此需要将这个函数作为子函数内部化，并且在子函数后面也要加上保存语句，注意在保存主函数时需要加上'-append'。
计算示例：
%% Example 2: MATLAB Functions
X = randn(100,3); Y = (X*[1 2 3]'+randn(100,1))>0;
profile on
[Beta_draws,ME1,ME2] = exampleFun(Y,X);
profileStruct = profile('info');
[flopTotal,Details] = FLOPS('exampleFun','exampleFunMat',profileStruct);这篇关于Matlab中使用FLOPS计算代码复杂度（浮点数运算次数）方法的文章就介绍到这儿，希望我们推荐的文章对编程师们有所帮助！
