Label posterior(const Eigen::DenseBase<Derived>& observations,
bool normalize = true) const {
assert(observations.cols() == num_features());
Label result = prior_;
for (std::size_t i = 0; i < num_features(); ++i) {
result *= feature(i).restrict_head(observations.col(i));
if (normalize) {
result.normalize();
}
}
return result;
}
// The number of training examples should always be a positive integer.
int DataNormalized::FeatureNormalize() {
const int kNumFeatures = num_features();
assert(kNumFeatures >= 1);
const int kNumTrainEx = num_train_ex();
assert(kNumTrainEx >= 1);
// Does not include dummy feature when normalizing.
const arma::mat kTrainingFeaturesNoDummy = \
training_features().cols(1,kNumFeatures);
const arma::vec mu_vec = \
arma::mean(kTrainingFeaturesNoDummy.cols(0,kNumFeatures-1)).t();
set_mu_vec(mu_vec);
const arma::vec sigma_vec = \
arma::stddev(kTrainingFeaturesNoDummy.cols(0,kNumFeatures-1)).t();
set_sigma_vec(sigma_vec);
arma::mat kTrainingFeaturesNoDummyNormalized = \
arma::zeros<arma::mat>(kNumTrainEx,kNumFeatures);
for(int row_index=0; row_index<kNumTrainEx;row_index++) {
kTrainingFeaturesNoDummyNormalized.row(row_index) = \
(kTrainingFeaturesNoDummy.row(row_index)-mu_vec.t())/sigma_vec.t();
}
// Appends dummy feature to normalized data, since normalized data will be
// used for linear regression.
const arma::mat kTrainingFeaturesNormalized = \
arma::join_horiz(arma::ones<arma::vec>(kNumTrainEx),\
kTrainingFeaturesNoDummyNormalized);
set_training_features_normalized(kTrainingFeaturesNormalized);
return 0;
}
// The number of training features should always be a positive integer.
// The number of training examples should always be a positive integer.
int DataUnlabeledNormalized::FeatureNormalize() {
const int kNumFeatures = num_features();
assert(kNumFeatures >= 1);
const int kNumTrainEx = num_train_ex();
assert(kNumTrainEx >= 1);
const arma::vec mu_vec = \
arma::mean(training_features().cols(0,kNumFeatures-1)).t();
set_mu_vec(mu_vec);
const arma::vec sigma_vec = \
arma::stddev(training_features().cols(0,kNumFeatures-1)).t();
set_sigma_vec(sigma_vec);
arma::mat kTrainingFeaturesNormalized = \
arma::zeros<arma::mat>(kNumTrainEx,kNumFeatures);
for(int row_index=0; row_index<kNumTrainEx;row_index++) {
kTrainingFeaturesNormalized.row(row_index) = \
(training_features().row(row_index)-mu_vec.t())/sigma_vec.t();
}
set_training_features_normalized(kTrainingFeaturesNormalized);
return 0;
}
