apaam/mlpack
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Astyled.

Browse Source
This commit is contained in:
Dongryeol Lee
2010-08-12 14:45:42 +00:00
parent 67db0b0480
commit 153fdc7124
2 changed files with 32 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib/trunk/contrib/dongryel/clusterwise_regression/clusterwise_regression.h
+2 -2
View File
@@ -41,7 +41,7 @@ class ClusterwiseRegressionResult {
private:
double Diameter_(const Matrix &dataset) const;
public:
const Matrix &coefficients() const;
@@ -74,7 +74,7 @@ class ClusterwiseRegression {
void MStep_(ClusterwiseRegressionResult &result_out);
void Solve_(int cluster_number, Vector *solution_out);
void UpdateMixture_(int cluster_number);
public:
fastlib/trunk/contrib/dongryel/clusterwise_regression/clusterwise_regression_dev.h
+30 -30
View File
@@ -17,9 +17,9 @@ const Matrix &ClusterwiseRegressionResult::coefficients() const {
return coefficients_;
}
Matrix &ClusterwiseRegressionResult::coefficients() {
return coefficients_;
}
Matrix &ClusterwiseRegressionResult::coefficients() {
return coefficients_;
}
double ClusterwiseRegressionResult::Predict(
const Vector &datapoint, int cluster_number) const {
@@ -45,7 +45,7 @@ double ClusterwiseRegressionResult::Predict(
const Vector &datapoint) const {
double mixed_prediction = 0;
for (int i = 0; i < mixture_weights_.length(); i++) {
for(int i = 0; i < mixture_weights_.length(); i++) {
mixed_prediction += mixture_weights_[i] *
Predict(datapoint, cluster_number);
}
@@ -67,13 +67,13 @@ ClusterwiseRegressionResult::ClusterwiseRegressionResult() {
double ClusterwiseRegressionResult::Diameter_(const Matrix &dataset_in) const {
std::vector< DRange > ranges;
ranges.resize(dataset_in.n_rows());
for (int j = 0; j < dataset_in.n_cols(); j++) {
for (int i = 0; i < dataset_in.n_rows(); i++) {
for(int j = 0; j < dataset_in.n_cols(); j++) {
for(int i = 0; i < dataset_in.n_rows(); i++) {
ranges[i] |= dataset_in.get(i, j);
}
}
double diameter = 0;
for (int i = 0; i < ranges.size(); i++) {
for(int i = 0; i < ranges.size(); i++) {
diameter += math::Sqr(ranges[i].width());
}
diameter = sqrt(diameter);
@@ -100,7 +100,7 @@ void ClusterwiseRegressionResult::Init(
// Initialize the bandwidths.
double diameter = Diameter_(*dataset_in);
for (int i = 0; i < num_clusters_in; i++) {
for(int i = 0; i < num_clusters_in; i++) {
kernels_[i].Init(math::Random(0.1 * diameter, 0.5 * diameter));
}
}
@@ -109,13 +109,13 @@ void ClusterwiseRegressionResult::Init(
namespace ml {
void ClusterwiseRegression::Solve_(int cluster_number, Vector *solution_out) {
// The right hand side weighted by the probabilities.
Vector right_hand_side;
right_hand_side.Init(targets_.length());
for(int i = 0; i < targets_.length(); i++) {
weighted_targets_[i] = targets_[i] * sqrt(
membership_probabilities_.get(i, cluster_number));
membership_probabilities_.get(i, cluster_number));
}
// The left hand side weighted by the probabilities.
@@ -126,8 +126,8 @@ void ClusterwiseRegression::Solve_(int cluster_number, Vector *solution_out) {
dataset_->MakeColumnVector(i, &point);
for(int j = 0; j < dataset_->n_rows(); j++) {
weighted_left_hand_side.set(
i, j, point[j] *
sqrt(membership_probabilities_.get(i, cluster_number)));
i, j, point[j] *
sqrt(membership_probabilities_.get(i, cluster_number)));
}
weighted_left_hand_side.set(
i, dataset_->n_rows(),
@@ -139,31 +139,31 @@ void ClusterwiseRegression::Solve_(int cluster_number, Vector *solution_out) {
la::QRInit(weighted_left_hand_side, &q_factor, &r_factor);
Vector q_trans_right_hand_side;
la::MulTransAInit(q_factor, right_hand_side, &q_trans_right_hand_side);
// SVD the R factor and solve it.
Matrix singular_values, left_singular_vectors,
right_singular_vectors_transposed;
Matrix singular_values, left_singular_vectors,
right_singular_vectors_transposed;
la::SVDInit(
r_factor, &singular_values, &left_singular_vectors,
&right_singular_vectors_transposed);
solution_out->SetZero();
for(int j = 0; j < singular_values.length(); j++) {
for(int j = 0; j < singular_values.length(); j++) {
if(singular_values[j] > 1e-6) {
Vector left_singular_vector;
left_singular_vectors.MakeColumnVector(j, &left_singular_vector);
double scaling_factor = la::Dot(
left_singular_vector, q_trans_right_hand_side) / singular_values[j];
left_singular_vector, q_trans_right_hand_side) / singular_values[j];
for(int k = 0; k < right_singular_vectors_transposed.length(); k++) {
(*solution_out)[k] += scaling_factor *
right_singular_vectors_transposed.get(j, k);
(*solution_out)[k] += scaling_factor *
right_singular_vectors_transposed.get(j, k);
}
}
}
}
void ClusterwiseRegression::UpdateMixture_(int cluster_number) {
}
void ClusterwiseRegression::EStep_(ClusterwiseRegressionResult &result_out) {
@@ -171,8 +171,8 @@ void ClusterwiseRegression::EStep_(ClusterwiseRegressionResult &result_out) {
// Compute the MLE of the posterior probability for each point for
// each cluster.
Vector probabilities_per_point;
probabilities_per_point.Init( kernels_.size() );
for (int k = 0; k < dataset_->n_cols(); k++) {
probabilities_per_point.Init(kernels_.size());
for(int k = 0; k < dataset_->n_cols(); k++) {
// Get the point.
Vector point;
@@ -180,12 +180,12 @@ void ClusterwiseRegression::EStep_(ClusterwiseRegressionResult &result_out) {
doiuble normalization = 0;
// Compute the probability for each cluster.
for (int j = 0; j < kernels_.size(); j++) {
for(int j = 0; j < kernels_.size(); j++) {
double squared_error;
double prediction = Predict(point, targets_[k], j, &squared_error);
probabilities_per_point[j] = mixture_weights_[j] *
kernels_[j].EvalUnnormOnSq(squared_error) /
kernels_[j].CalcNormConstant( point.length() );
probabilities_per_point[j] = mixture_weights_[j] *
kernels_[j].EvalUnnormOnSq(squared_error) /
kernels_[j].CalcNormConstant(point.length());
normalization += probabilities_per_point[j];
} // end of looping through each cluster.
@@ -193,7 +193,7 @@ void ClusterwiseRegression::EStep_(ClusterwiseRegressionResult &result_out) {
membership_probabilities_.set(
k, j, probabilities_per_point[j] / normalization);
} // end of looping through each cluster.
} // end of looping through each point.
}
@@ -205,9 +205,9 @@ void ClusterwiseRegression::MStep_(ClusterwiseRegressionResult &result_out) {
for(int j = 0; j < dataset_->n_cols(); j++) {
sum += membereship_probabilities_.get(j, i);
}
mixture_weights_[i] = sum / ((double) dataset_->n_cols());
mixture_weights_[i] = sum / ((double) dataset_->n_cols());
}
// Update the linear model for each cluster.
Matrix &coefficients = result_out.coefficients();
for(int j = 0; j < kernels_.size(); j++) {
@@ -247,7 +247,7 @@ void ClusterwiseRegression::Compute(
MStep_(*result_out);
}
while (Converged_() == false);
while(Converged_() == false);
}
};