void test_predict(){
crbm m;
dataset_blas train_set;
int i, j, k;
int minibatch = 20;
int batch_size, niter;
double *x, *y = py;
load_model(&m, "../data/20newsgroup/cross_valid/crbm.model");
load_corpus("../data/20newsgroup/train.data.format", &train_set);
load_corpus_label("../data/20newsgroup/train.label.format", &train_set);
niter = ceil(train_set.N * 1.0 / minibatch);
for(k = 0; k < niter; k++){
if(k == niter - 1){
batch_size = train_set.N - minibatch * (niter-1);
}else{
batch_size = minibatch;
}
x = train_set.input + m.nvisible * minibatch * k;
get_y_given_x(&m, x, y, batch_size);
for(j = 0; j < batch_size; j++){
printf("%d : %d\n", train_set.output[k*minibatch+j] + 1,
get_max_index(&y[j*m.ncat], m.ncat));
}
}
free_crbm(&m);
free_dataset_blas(&train_set);
}
int main() {
std::clock_t start,end;
start = std::clock();
Node root = load_corpus();
end = std::clock();
// std::cout << "Time: " << (end-start)/(double)(CLOCKS_PER_SEC/1000) << "\n";
std::string prefix;
std::cout << "Enter a prefix: ";
std::getline(std::cin,prefix);
std::vector <std::string> suggestedWords = root.search(prefix);
std::sort(suggestedWords.begin(),suggestedWords.end(),compareRanks);
int num_words = suggestedWords.size();
for(int i = 0; i < num_words; i++) {
std::string suggestedWord = suggestedWords[i];
std::cout << "Suggested: " << suggestedWord << ", #" << wordRank[suggestedWord] << std::endl;
//std::cout << "Suggested: "<<root.search(prefix)[i] << std::endl;
}
}
void test_crbm(){
dataset_blas train_set, valid_set;
int nhidden = 500;
int epoch = 30;
double lr = 0.1;
int minibatch = 10;
double momentum = 0;
load_corpus("../data/20newsgroup/cross_valid/data.out", &train_set);
load_corpus_label("../data/20newsgroup/cross_valid/label.out", &train_set);
//load_corpus("../data/20newsgroup/train.data.format", &train_set);
//load_corpus_label("../data/20newsgroup/train.label.format", &train_set);
//load_corpus("../data/tcga/train.pm.data", &train_set);
//load_corpus_label("../data/tcga/train.pm.label", &train_set);
//load_mnist_dataset_blas(&train_set, &valid_set);
train_crbm(&train_set, train_set.n_feature, nhidden, train_set.nlabel, epoch, lr,
minibatch, momentum, "../data/20newsgroup/cross_valid/crbm.model");
free_dataset_blas(&train_set);
}
