int
CNetComm::safe_net_out (const char *sndbuf, int sndlen)
{
DEBUG_LOG("enter the safe_net_out ....");
int len = net_out(sndbuf, sndlen);
if (len == -1) {
DEBUG_LOG("reconnect the server ...");
if (open_conn() == -1) {
close_conn();
DEBUG_LOG("reconnect the server fail !");
return -1;
}
return net_out(sndbuf, sndlen); //send the data second time
}
return len;
}
// compute the features of specified charsamp and
// feedforward the specified nets
bool HybridNeuralNetCharClassifier::RunNets(CharSamp *char_samp) {
int feat_cnt = feat_extract_->FeatureCnt();
int class_cnt = char_set_->ClassCount();
// allocate i/p and o/p buffers if needed
if (net_input_ == NULL) {
net_input_ = new float[feat_cnt];
if (net_input_ == NULL) {
return false;
}
net_output_ = new float[class_cnt];
if (net_output_ == NULL) {
return false;
}
}
// compute input features
if (feat_extract_->ComputeFeatures(char_samp, net_input_) == false) {
return false;
}
// go thru all the nets
memset(net_output_, 0, class_cnt * sizeof(*net_output_));
float *inputs = net_input_;
for (int net_idx = 0; net_idx < nets_.size(); net_idx++) {
// run each net
vector<float> net_out(class_cnt, 0.0);
if (!nets_[net_idx]->FeedForward(inputs, &net_out[0])) {
return false;
}
// add the output values
for (int class_idx = 0; class_idx < class_cnt; class_idx++) {
net_output_[class_idx] += (net_out[class_idx] * net_wgts_[net_idx]);
}
// increment inputs pointer
inputs += nets_[net_idx]->in_cnt();
}
Fold();
return true;
}
