新建 文本文档 (2).txt

此程序用于测试神经元的库函数 // 它展示了怎样用这个库所生成的多层BP网络来求解宇称问题

//
// demo2.cpp
// 2004-05-24 by ShanLanshan
//
// 此程序用于测试神经元的库函数
// 它展示了怎样用这个库所生成的多层BP网络来求解宇称问题
// 在Visual Stduio 2003上编译通过，要注意启动运行时信息的支持

#include <stdio.h>
#include <assert.h>
#include <errno.h>
#include "public/ann/neuron.h"
#include "public/ann/bp/bpneuron.h"
#include "public/prob.h"
#include "public/ann/annlib.h"
#include "public/ann/vec_container.h"
#include "public/ann/vector.h"
#include "public/ann/annfunc/annfunc.h"
#include "Windows.h"
#include <fstream>

using namespace ann;

int main()
{
    char s[1024];
    const char *file = "C:\\bpnet.txt";

    for (int i=0; i< 0; ++i) {
        std::cout << prob_get_range(0, 3) << " ";
    }
    //std::cin >> s;

    AnnLib annlib;
    if (!annlib.init()) {
        std::cout << "ann lib init failed\n";
        return -1;
    }

    AnnFuncGroup *fg;
    fg = annfunc_get("sign");
    assert(fg != NULL);

    AnnSampleSet ass("宇称5");

    BPNet net;
    AnnVector vec_input, vec_output, vec_tutor;

    bool openfile = true;
    std::ifstream is;
    if (openfile)
        is.open(file);
    if (openfile && is.is_open()) {
        std::cout << "open file ok\n";
        net << is;
    } else {
        char initstr[64];
        sprintf(initstr, "%d * %s * %d",
            ass.m_input_division,
            "100 * 16",
            ass.m_output_division);
        std::cout << "Init: " << initstr << std::endl;
        net.init(initstr);
        net.mutual_conn();
        prob_set_seed(0);
        net.random();
    } 

    //net.set_output_func(annfunc_get("sech6"), 0);
    net.set_output_func(annfunc_get("sech6"), 1);
    net.set_output_func(annfunc_get("sin"), 2);
    //net.set_output_func(annfunc_get("sech6"), 3);
    net.set_output_func(annfunc_get("sech6_cache"));
    //net.set_output_func(annfunc_get("sech6"));
    //net.set_output_func(annfunc_get("sin"));

    vec_output.set_size(net.get_output_division());

    int right = 0, k = 0;
    time_t t1,t2;

    t1 = time(NULL);

    for (int m = 0; m < 1; ++m) {
        //net.random();
        right = 0;
        std::cout << m << std::endl;
        for (int i=400000; i>0; --i) {
            printf(".");
            ass.get_sample(&vec_input, &vec_tutor);
            net.set_input(&vec_input);
            net.convert();
            net.get_output(&vec_output);
            net.learn(&vec_tutor);
            fg->multi_main(vec_output.get_buffer(), vec_output.get_size(), NULL);
            vec_output.release_buffer();

            if (0 && (i % 100) == 0) {
                std::cout << i << std::endl;
                Sleep(10);
            }
            //if (i < 200)
            if (0) 
                std::cout << "\n\nSample" << i << " " << vec_input << std::endl
                << "tutor : " << vec_tutor << std::endl
                << "output: " << vec_output << std::endl
                << std::endl;
            k ++;
            if (vec_tutor == vec_output) {
                ++right;
                if (right >= 600) break;
            } else {
                right = 0;
            }
        }
    }

    t2 = time(NULL);
    std::cout << static_cast<long>(t2 - t1) << "seconds passed\n";
    std::cout << "Get correct in step " << k << ", ok\n";

    std::ofstream fs;
    fs.open(file);
    fs << net;

    std::cin >> s;
    return 0;
}