void DeferredRender::doRender() {
if(!loaded)
return;
//std::cout << "geometry" << std::endl;
geometryPass();
//backgroundPass();
//std::cout << "light" << std::endl;
lightPass();
//std::cout << "alpha" << std::endl;
alphaPass();
// TODO: ajouter pass SSAO, MXAA
//std::cout << "final" << std::endl;
renderScreen();
//throw -1;
}
int main(int argc, const char *argv[])
{
int N,
M,
T,
maxIters,
seed,
i,
j,
iter,
str_len;
char **alphabet;
double logProb,
newLogProb;
double *pi,
*piBar,
**A,
**Abar,
**B,
**Bbar;
struct stepStruct *step;
FILE *in,
*out;
char s[80];
int wantTraining = 1;
if(argc != 10)
{
fprintf(stderr, "\nUsage: %s N M T maxIters filename alphabet modelfile seed\n\n", argv[0]);
fprintf(stderr, "where N == number of states of the HMM\n");
fprintf(stderr, " M == number of observation symbols\n");
fprintf(stderr, " T == number of observations in the training set\n");
fprintf(stderr, " maxIters == max iterations of re-estimation algorithm\n");
fprintf(stderr, " filename == name of input file\n");
fprintf(stderr, " alphabet == name of file defining the alphabet\n");
fprintf(stderr, " modelfile == name of model output file\n");
fprintf(stderr, " seed == seed value for pseudo-random number generator (PRNG)\n\n");
fprintf(stderr, " wantTraining == to train enter 1, otherwise 0 \n\n");
fprintf(stderr, "For example:\n\n %s 2 10 10000 500 datafile alphabet modelfile 1241\n\n", argv[0]);
fprintf(stderr, "will create a HMM with 2 states and 10 observation symbols,\n");
fprintf(stderr, "will read in the first 10000 observations from `datafile',\n");
fprintf(stderr, "will use the observation symbols defined in file `alphabet', and\n");
fprintf(stderr, "will write the model (pi, A, B) to `modelfile', and\n");
fprintf(stderr, "will seed the PRNG with 1241 and train the HMM with a maximum of 500 iterations.\n\n");
exit(0);
}
N = atoi(argv[1]);
M = atoi(argv[2]);
T = atoi(argv[3]);
maxIters = atoi(argv[4]);
seed = atoi(argv[8]);
wantTraining = atoi(argv[9]);
pi = (double *)malloc(N * sizeof(double));
piBar = (double *)malloc(N * sizeof(double));
A = (double **)malloc(N * sizeof(double*));
Abar =static_cast<double **>(malloc(N * sizeof(double*)));
for (i=0; i<N; ++i)
{
A[i] = static_cast<double *>(malloc(N * sizeof(double)));
Abar[i] = static_cast<double *>(malloc(N * sizeof(double)));
}
B = static_cast<double **>(malloc(N * sizeof(double*)));
Bbar = static_cast<double **>(malloc(N * sizeof(double*)));
for (i=0; i<N; ++i)
{
B[i] = static_cast<double *>(malloc(M * sizeof(double)));
Bbar[i] = static_cast<double *>(malloc(M * sizeof(double)));
}
////////////////////////
// read the data file //
////////////////////////
// allocate memory
printf("allocating %d bytes of memory... ", (T + 1) * sizeof(struct stepStruct));
fflush(stdout);
if((step = static_cast<stepStruct *>(calloc(T + 1, sizeof(struct stepStruct)))) == NULL)
{
fprintf(stderr, "\nUnable to allocate alpha\n\n");
exit(0);
}
for (i=0; i<T+1; ++i)
{
step[i].alpha = static_cast<double *>(malloc(N * sizeof(double)));
step[i].beta = static_cast<double *>(malloc(N * sizeof(double)));
step[i].gamma = static_cast<double *>(malloc(N * sizeof(double)));
step[i].diGamma = static_cast<double **>(malloc(N * sizeof(double*)));
for (j=0; j<N; ++j)
{
step[i].diGamma[j] = static_cast<double *>(malloc(N * sizeof(double)));
}
}
printf("done\n");
// read in the observations from file
printf("GetObservations... ");
fflush(stdout);
in = fopen(argv[5], "r"); // argv[5] = filename
if(in == NULL)
{
fprintf(stderr, "\nError opening file %s\n\n", argv[5]);
exit(0);
}
i = 0;
fgets(s,80,in); // get rid of the first line
while (i < T)
{
fgets(s,80,in);
step[i].obs = atoi(s);
++i;
}
fclose(in);
printf("done\n");
// read in the alphabet from file
printf("GetAlphabet... ");
fflush(stdout);
alphabet = static_cast<char **>(malloc(M * sizeof (char*)));
in = fopen(argv[6], "r"); // argv[6] = alphabet
if(in == NULL)
{
fprintf(stderr, "\nError opening file %s\n\n", argv[6]);
exit(0);
}
i = 0;
fgets(s,80,in); // get rid of the first line
while (i < M)
{
fgets(s,80,in);
str_len = strlen(s);
alphabet[i] = static_cast<char *>(malloc(str_len * sizeof(char)));
strncpy(alphabet[i], s, str_len-1);
alphabet[i][str_len-1] = '\0';
++i;
}
fclose(in);
printf("done\n");
/////////////////////////
// hidden markov model //
/////////////////////////
srand(seed);
// initialize pi[], A[][] and B[][]
initMatrices(pi, A, B, N, M, seed);
// print pi[], A[][] and B[][] transpose
printf("\nN = %d, M = %d, T = %d\n", N, M, T);
printf("initial pi =\n");
printPi(pi, N);
printf("initial A =\n");
printA(A, N);
printf("initial B^T =\n");
printBT(B, N, M, alphabet);
// initialization
iter = 0;
logProb = -1.0;
newLogProb = 0.0;
if (wantTraining) {
// main loop
while((iter < maxIters) && (newLogProb > logProb))
{
printf("\nbegin iteration = %d\n", iter);
logProb = newLogProb;
// alpha (or forward) pass
printf("alpha pass... ");
fflush(stdout);
alphaPass(step, pi, A, B, N, T);
printf("done\n");
// beta (or backwards) pass
printf("beta pass... ");
fflush(stdout);
betaPass(step, pi, A, B, N, T);
printf("done\n");
// compute gamma's and diGamma's
printf("compute gamma's and diGamma's... ");
fflush(stdout);
computeGammas(step, pi, A, B, N, T);
printf("done\n");
// find piBar, reestimate of pi
printf("reestimate pi... ");
fflush(stdout);
reestimatePi(step, piBar, N);
printf("done\n");
// find Abar, reestimate of A
printf("reestimate A... ");
fflush(stdout);
reestimateA(step, Abar, N, T);
printf("done\n");
// find Bbar, reestimate of B
printf("reestimate B... ");
fflush(stdout);
reestimateB(step, Bbar, N, M, T);
printf("done\n");
#ifdef PRINT_REESTIMATES
printf("piBar =\n");
printPi(piBar, N);
printf("Abar =\n");
printA(Abar, N);
printf("Bbar^T = \n");
printBT(Bbar, N, M, alphabet);
#endif // PRINT_REESTIMATES
// assign pi, A and B corresponding "bar" values
for(i = 0; i < N; ++i)
{
pi[i] = piBar[i];
for(j = 0; j < N; ++j)
{
A[i][j] = Abar[i][j];
}
for(j = 0; j < M; ++j)
{
B[i][j] = Bbar[i][j];
}
}// next i
// compute log [P(observations | lambda)], where lambda = (A,B,pi)
newLogProb = 0.0;
for(i = 0; i < T; ++i)
{
newLogProb += log(step[i].c);
}
newLogProb = -newLogProb;
// a little trick so that no initial logProb is required
if(iter == 0)
{
logProb = newLogProb - 1.0;
}
printf("completed iteration = %d, log [P(observation | lambda)] = %f\n",
iter, newLogProb);
++iter;
}// end while
out = fopen(argv[7], "w"); // argv[7] = modelfile
writeModel(pi, A, B, N, M, T, alphabet, out);
fclose(out);
printf("\nT = %d, N = %d, M = %d, iterations = %d\n\n", T, N, M, iter);
printf("final pi =\n");
printPi(pi, N);
printf("\nfinal A =\n");
printA(A, N);
printf("\nfinal B^T =\n");
printBT(B, N, M, alphabet);
printf("\nlog [P(observations | lambda)] = %f\n\n", newLogProb);
} // end of training
else { //want to do testing
out = fopen(argv[7], "r"); // argv[7] = modelfile
readModelFile(pi, A, B, N, M, T, alphabet, out);
// alpha (or forward) pass
printf("alpha pass... ");
fflush(stdout);
alphaPass(step, pi, A, B, N, T);
printf("done\n");
printf("logProb %f\n", computeLogProb(step, T)/T);
// FILE * newFile = fopen("testing.txt", "a");
//writeModel(pi, A, B, N, M, T, alphabet, newFile);
//fclose(newFile);
fclose(out);
} // end of testing
}// end hmm
