static void test_iso(const psl_ctx_t *psl, const char *domain, const char *expected_result)
{
/* makes only sense with a runtime IDN library configured */
#if defined(WITH_LIBIDN) || defined(WITH_LIBIDN2) || defined(WITH_LIBICU)
testx(psl, domain, "iso-8859-15", "de", expected_result);
#endif
}
double cross_validation(Matrix<double> x, Vector<double> y, Kernel *k, double c, int cross = -1)
{
// 交差検定
int group = cross;
if (group <= 0) group = 1 + log2(x.nrows());
int size = x.nrows() / group;
double accuracy = 0.0;
for (int i = 0; i < group; i++) {
size_t si = i * size, sz = size + (i == group - 1 ? (x.nrows() % size) : 0);
Matrix<double> samplex(x.nrows() - sz, x.ncols()), testx(sz, x.ncols());
Vector<double> sampley(y.size() - sz), testy(sz);
for (size_t j = 0; j < x.nrows(); j++) {
if (j < si) {
samplex.setRow(j, x.extractRow(j));
sampley[j] = y[j];
}
else if (j < si + sz) {
testx.setRow(j - si, x.extractRow(j));
testy[j - si] = y[j];
}
else {
samplex.setRow(j - sz, x.extractRow(j));
sampley[j - sz] = y[j];
}
}
SVM svm(samplex, sampley, k, c);
int pass = 0;
for (size_t j = 0; j < sz; j++)
if (svm.discriminant(testx.extractRow(j)) * testy[j] > 0.0) pass++;
double acc = pass / (double)sz * 100.0;
printf("loop #%d :: %f\n", i, acc);
accuracy += acc / group;
}
printf("total accuracy :: %f\n", accuracy);
return accuracy;
}
static void test(const psl_ctx_t *psl, const char *domain, const char *expected_result)
{
testx(psl, domain, "utf-8", "en", expected_result);
}
int run_sampler(unsigned epochs,float alpha,unsigned batch_size){
//load sampler
string training_path = "PPAttachData/training.lemma";
string param_path = "PPAttachData/wordsketches/";
string vpath = param_path + string("vdistrib");
string x1vpath = param_path + string("x1givenv");
string pvpath = param_path + string("pgivenv");
string x2vppath = param_path + string("x2givenvp");
string px1path = param_path + string("pgivenx1");
string x2x1ppath = param_path + string("x2givenx1p");
DataSampler samp(training_path.c_str(),
vpath.c_str(),
x1vpath.c_str(),
pvpath.c_str(),
x2vppath.c_str(),
px1path.c_str(),
x2x1ppath.c_str());
//load dev and test
PPADataEncoder dev_set("PPAttachData/devset.lemma");
PPADataEncoder test_set("PPAttachData/test.lemma");
//load Word vectors
Word2vec w2v;
vector<string> wvdict;
af::array w2v_embeddings;
w2v.load_dictionary("PPAttachData/embeddings/deps.words.lemmatized");
//w2v.filter(xdict);
//make network
vector<string> ydict;
samp.getYdictionary(ydict);
SymbolicFeedForwardNetwork<string,string> net;
net.set_output_layer("loss",new SoftMaxLoss<string>(ydict));
net.add_layer("top",new LinearLayer());
net.add_layer("hidden",new ReLUActivation(400));
net.add_layer("A",new LinearLayer());
net.add_input_layer("lookupA",new LinearLookup<string>(w2v.get_keys(),w2v.get_values(),4,false));
net.connect_layers("loss","top");
net.connect_layers("top","hidden");
net.connect_layers("hidden","A");
net.connect_layers("A","lookupA");
for(int E = 0; E < epochs;++E){
vector<string> ydata;
vector<vector<string>> xdata;
//af::timer start1 = af::timer::start();
samp.generate_sample(ydata,xdata,batch_size);
//printf("elapsed seconds (sampling): %g\n", af::timer::stop(start1));
PPADataEncoder sampdata(ydata,xdata);
vector<string> enc_ydata;
vector<vector<string>> enc_xdata(1,vector<string>());
sampdata.getYdata(enc_ydata);
sampdata.getXdata(enc_xdata[0]);
//af::timer start2 = af::timer::start();
net.set_batch_data(enc_ydata,enc_xdata);
float loss = net.train_one(alpha,true,true);
//printf("elapsed seconds (backprop): %g\n", af::timer::stop(start2));
if (E % 20 == 0){
vector<string> devy;
vector<vector<string>> devx(1,vector<string>());
dev_set.getYdata(devy);
dev_set.getXdata(devx[0]);
float acc = net.eval_avg(devy,devx); //auto-eval on dev data
cout << "epoch " << E << ", loss= " << loss << ", eval (dev) = " << acc << endl;
}else {
cout << "epoch" << E <<endl;
}
}
vector<string> testy;
vector<vector<string>> testx(1,vector<string>());
test_set.getYdata(testy);
test_set.getXdata(testx[0]);
float acc = net.eval_avg(testy,testx);
cout << "final eval (test) = " << acc << endl;
return 0;
}
