davidalbertonogueira / mlp Goto Github PK

Am I missing something or the biases are not considered?

Check example: easylogging/easyloggingpp/samples/STL/configurator.cpp :

#include "easylogging++.h"

INITIALIZE_EASYLOGGINGPP

int main(int argc, char** argv) {
    START_EASYLOGGINGPP(argc, argv);
    LOG(INFO) << "Info log using 'default' logger before using configuration";
   
    el::Configurations confFromFile("../default-logger.conf");
 
    el::Loggers::reconfigureAllLoggers(confFromFile); 
     
    LOG(INFO) << "Info log after manually configuring 'default' logger";
    el::Loggers::getLogger("default")->reconfigure();
    LOG(ERROR) << "Error log";
    LOG(WARNING) << "WARNING! log";
    VLOG(1) << "Verbose log 1";
    VLOG(2) << "Verbose log 2";
    return 0;
}

Hello;

I've downloaded your project and I am trying to build the example: I've found that in the file "Utils.h" appears a reference to a include file called Crono.h that doesn't appear.

Hello (sorry my English), i wont create neural web with 3 hidden layers, i am write this: (3 input, two layers on 3 neurons, and 2 output).

mWeb = new MLP({ 3, 3, 3, 2 }, { "sigmoid", "linear" }, false);
std::vector<TrainingSample> traning_sample = {
		{{1, 0, 0}, {1, 0}},
		{{0, 1, 0}, {1, 1}},
		{{0, 1, 1}, {0, 0}}
};
mWeb->Train(traning_sample, 0.01, 5000, 0.10, false);

it's correct? i have errors with this code (time out).
if it is not correct, please, write me correct realization.

if i am write this:

mWeb = new MLP({ 3, 3, 2 }, { "sigmoid", "linear" }, false);
std::vector<TrainingSample> traning_sample = {
		{{1, 0, 0}, {1, 0}},
		{{0, 1, 0}, {1, 1}},
		{{0, 1, 1}, {0, 0}}
};
mWeb->Train(traning_sample, 0.01, 5000, 0.10, false);

is works ok.

Does derivative of softmax considered during backward pass in your code?

Thanks!

I tried to compile the project, but make fails with the following error:

Compiling program NodeTest g++ src/NodeTest.o src/MLP.o -std=gnu++11 -std=c++11 -Wall -fmessage-length=0 -fPIC -I./deps -I -g3 -L -v -o NodeTest ld: warning: directory not found for option '-L-v' Creating static lib mlp.a ar rcs mlp.a ./src/MLP.o Creating dynamic lib mlp.so g++ -o mlp.so ./src/MLP.o -shared -L -v ld: warning: directory not found for option '-L-v' Undefined symbols for architecture x86_64: "el::base::utils::s_currentHost", referenced from: el::base::LogFormat::updateFormatSpec() in MLP.o "el::base::utils::s_currentUser", referenced from: el::base::LogFormat::updateFormatSpec() in MLP.o "el::base::elStorage", referenced from: el::base::Writer::construct(int, char const*, ...) in MLP.o el::base::Writer::initializeLogger(std::__1::basic_string<char, std::__1::char_traits<char>, std::__1::allocator<char> > const&, bool, bool) in MLP.o el::base::MessageBuilder::initialize(el::Logger*) in MLP.o el::base::MessageBuilder::operator<<(char const*) in MLP.o el::base::Writer::processDispatch() in MLP.o el::base::Writer::triggerDispatch() in MLP.o el::base::LogDispatcher::dispatch() in MLP.o ... ld: symbol(s) not found for architecture x86_64 clang: error: linker command failed with exit code 1 (use -v to see invocation) make: *** [mlp.so] Error 1

any idea what might be? maybe a required missing dependency?

Thanks.

In your example you use "sigmoid" and "linear".

Are there any other type of function that I can use?

Thanks!

Does your MLP model do regression tasks too? or is it just a Classifier?