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

Naive algorithm now uses much less memory

Browse Source
This commit is contained in:
Dongryeol Lee
2008-02-20 01:19:15 +00:00
parent 3eac4327e9
commit ac0831b7dd
2 changed files with 36 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files
fastlib2/contrib/dongryel/regression/dense_lpr_impl.h
+6 -2
View File
@@ -59,7 +59,9 @@ void DenseLpr<TKernel, TPruneRule>::BasicComputeDualTree_
// Initialize storage for the final results.
query_regression_estimates->Init(queries.n_cols());
query_magnitude_weight_diagrams->Init(queries.n_cols());
query_influence_values->Init(queries.n_cols());
if(query_influence_values != NULL) {
query_influence_values->Init(queries.n_cols());
}
// Three steps: initialize the query tree, then call dualtree,
// then final postprocess.
@@ -145,7 +147,9 @@ void DenseLpr<TKernel, TPruneRule>::BasicComputeSingleTree_
// Initialize storage for the final results.
query_regression_estimates->Init(queries.n_cols());
query_magnitude_weight_diagrams->Init(queries.n_cols());
query_influence_values->Init(queries.n_cols());
if(query_influence_values != NULL) {
query_influence_values->Init(queries.n_cols());
}
// iterate over each query point.
for(index_t q = 0; q < queries.n_cols(); q++) {
fastlib2/contrib/dongryel/regression/naive_lpr.h
+30 -38
View File
@@ -109,29 +109,31 @@ class NaiveLpr {
}
// Temporary variables to hold intermediate computation results.
ArrayList<Matrix> denominator;
denominator.Init(queries.n_cols());
Matrix numerator;
numerator.Init(total_num_coeffs_, queries.n_cols());
numerator.SetZero();
ArrayList<Matrix> weight_diagram_numerator;
weight_diagram_numerator.Init(queries.n_cols());
for(index_t i = 0; i < queries.n_cols(); i++) {
denominator[i].Init(total_num_coeffs_, total_num_coeffs_);
denominator[i].SetZero();
weight_diagram_numerator[i].Init(total_num_coeffs_, total_num_coeffs_);
weight_diagram_numerator[i].SetZero();
}
Matrix denominator;
denominator.Init(total_num_coeffs_, total_num_coeffs_);
Vector numerator;
numerator.Init(total_num_coeffs_);
Matrix weight_diagram_numerator;
weight_diagram_numerator.Init(total_num_coeffs_, total_num_coeffs_);
// Temporary variable for storing multivariate expansion of a
// point.
Vector point_expansion;
point_expansion.Init(total_num_coeffs_);
// Temporary variable for holding the pseudoinverse.
Matrix denominator_inv_q;
denominator_inv_q.Init(total_num_coeffs_, total_num_coeffs_);
// Compute unnormalized sum for the numerator vector and the
// denominator matrix.
for(index_t q = 0; q < queries.n_cols(); q++) {
// Initialize the temporary variables holding the sum.
numerator.SetZero();
denominator.SetZero();
weight_diagram_numerator.SetZero();
// Get the query point.
const double *q_col = queries.GetColumnPtr(q);
for(index_t r = 0; r < rset_.n_cols(); r++) {
@@ -152,16 +154,15 @@ class NaiveLpr {
for(index_t i = 0; i < total_num_coeffs_; i++) {
// Compute each component of the numerator matrix.
numerator.set(i, q, numerator.get(i, q) +
r_target * kernel_value * point_expansion[i]);
numerator[i] += r_target * kernel_value * point_expansion[i];
// Here, compute each component of the denominator matrix.
for(index_t j = 0; j < total_num_coeffs_; j++) {
denominator[q].set(j, i, denominator[q].get(j, i) +
point_expansion[j] * point_expansion[i] *
kernel_value);
weight_diagram_numerator[q].set
(j, i, weight_diagram_numerator[q].get(j, i) +
denominator.set(j, i, denominator.get(j, i) +
point_expansion[j] * point_expansion[i] *
kernel_value);
weight_diagram_numerator.set
(j, i, weight_diagram_numerator.get(j, i) +
point_expansion[j] * point_expansion[i] * kernel_value *
kernel_value);
@@ -171,18 +172,10 @@ class NaiveLpr {
// vector.
} // end of looping over each reference point
} // end of looping over each query point
Matrix denominator_inv_q;
denominator_inv_q.Init(total_num_coeffs_, total_num_coeffs_);
// now iterate over all query points and compute regression estimate
for(index_t q = 0; q < queries.n_cols(); q++) {
const double *q_col = queries.GetColumnPtr(q);
// The coefficients computed for the local fit at the given
// query point.
Vector beta_q;
Vector q_numerator;
numerator.MakeColumnVector(q, &q_numerator);
// Compute the query point expansion.
MultiIndexUtil::ComputePointMultivariatePolynomial
@@ -190,8 +183,8 @@ class NaiveLpr {
// Now invert the denominator matrix for each query point and
// multiply by the numerator vector.
MatrixUtil::PseudoInverse(denominator[q], &denominator_inv_q);
la::MulInit(denominator_inv_q, q_numerator, &beta_q);
MatrixUtil::PseudoInverse(denominator, &denominator_inv_q);
la::MulInit(denominator_inv_q, numerator, &beta_q);
// Compute the dot product between the multiindex vector for the
// query point by the beta_q.
@@ -199,14 +192,13 @@ class NaiveLpr {
// Now we compute the magnitude of the weight diagram for each
// query point.
Vector pseudo_inverse_times_query_expansion;
Vector intermediate_product;
Vector pseudo_inverse_times_query_expansion, intermediate_product;
la::MulInit(denominator_inv_q, point_expansion,
&pseudo_inverse_times_query_expansion);
la::MulInit(weight_diagram_numerator[q],
la::MulInit(weight_diagram_numerator,
pseudo_inverse_times_query_expansion, &intermediate_product);
(*query_magnitude_weight_diagrams)[q] =
sqrt(la::Dot(pseudo_inverse_times_query_expansion,
sqrt(la::Dot(pseudo_inverse_times_query_expansion,
intermediate_product));
// Compute the influence value at each point (if it belongs to
@@ -216,7 +208,7 @@ class NaiveLpr {
(*query_influence_values)[q] =
la::Dot(point_expansion, pseudo_inverse_times_query_expansion);
}
} // end of iterating over each query point.
} // end of looping over each query point
}
void ComputeConfidenceBands_(const Matrix &queries,