tommy/libpostal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[math] Generic dense matrix implementation using BLAS calls for matrix-matrix multiplication if available

Browse Source
This commit is contained in:
Al
2016-08-06 00:40:01 -04:00
parent d4a792f33c
commit 46cd725c13
19 changed files with 469 additions and 425 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
src/logistic_regression.c
View File

@@ -6,7 +6,7 @@
#define NEAR_ZERO_WEIGHT 1e-6
bool logistic_regression_model_expectation(matrix_t *theta, sparse_matrix_t *x, matrix_t *p_y) {
bool logistic_regression_model_expectation(double_matrix_t *theta, sparse_matrix_t *x, double_matrix_t *p_y) {
if (theta == NULL || x == NULL || p_y == NULL) return false;
if (sparse_matrix_dot_dense(x, theta, p_y) != 0) {
@@ -18,12 +18,12 @@ bool logistic_regression_model_expectation(matrix_t *theta, sparse_matrix_t *x,
return true;
}
double logistic_regression_cost_function(matrix_t *theta, sparse_matrix_t *x, uint32_array *y, matrix_t *p_y, double lambda) {
double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y, double lambda) {
size_t m = x->m;
size_t n = x->n;
if (m != y->n) return -1.0;
if (!matrix_resize(p_y, x->m, theta->n)) {
if (!double_matrix_resize(p_y, x->m, theta->n)) {
return -1.0;
}
@@ -36,7 +36,7 @@ double logistic_regression_cost_function(matrix_t *theta, sparse_matrix_t *x, ui
for (size_t i = 0; i < p_y->m; i++) {
uint32_t y_i = y->a[i];
double value = matrix_get(p_y, i, y_i);
double value = double_matrix_get(p_y, i, y_i);
if (value > NEAR_ZERO_WEIGHT) {
cost += log(value);
}
@@ -48,7 +48,7 @@ double logistic_regression_cost_function(matrix_t *theta, sparse_matrix_t *x, ui
double reg_cost = 0.0;
for (size_t i = 1; i < theta->m; i++) {
for (size_t j = 0; j < theta->n; j++) {
double theta_ij = matrix_get(theta, i, j);
double theta_ij = double_matrix_get(theta, i, j);
reg_cost += theta_ij * theta_ij;
}
@@ -61,13 +61,13 @@ double logistic_regression_cost_function(matrix_t *theta, sparse_matrix_t *x, ui
}
static bool logistic_regression_gradient_params(matrix_t *theta, matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, matrix_t *p_y,
static bool logistic_regression_gradient_params(double_matrix_t *theta, double_matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y,
uint32_array *x_cols, double lambda) {
size_t m = x->m;
size_t n = x->n;
if (m != y->n) return false;
if (!matrix_resize(p_y, x->m, theta->n) || !matrix_resize(p_y, x->m, theta->n)) {
if (!double_matrix_resize(p_y, x->m, theta->n) || !double_matrix_resize(p_y, x->m, theta->n)) {
return false;
}
@@ -100,11 +100,11 @@ static bool logistic_regression_gradient_params(matrix_t *theta, matrix_t *gradi
uint32_t *cols = x_cols->a;
for (i = 0; i < batch_rows; i++) {
col = cols[i];
gradient_i = matrix_get_row(gradient, col);
gradient_i = double_matrix_get_row(gradient, col);
double_array_zero(gradient_i, num_classes);
}
} else {
matrix_zero(gradient);
double_matrix_zero(gradient);
}
// gradient = -(1. / m) * x.T.dot(y - p_y) + lambda * theta
@@ -133,7 +133,7 @@ static bool logistic_regression_gradient_params(matrix_t *theta, matrix_t *gradi
}
}
} else {
matrix_mul(gradient, scale);
double_matrix_mul(gradient, scale);
}
@@ -166,12 +166,12 @@ static bool logistic_regression_gradient_params(matrix_t *theta, matrix_t *gradi
return true;
}
inline bool logistic_regression_gradient_sparse(matrix_t *theta, matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, matrix_t *p_y,
inline bool logistic_regression_gradient_sparse(double_matrix_t *theta, double_matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y,
uint32_array *x_cols, double lambda) {
return logistic_regression_gradient_params(theta, gradient, x, y, p_y, x_cols, lambda);
}
inline bool logistic_regression_gradient(matrix_t *theta, matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, matrix_t *p_y, double lambda) {
inline bool logistic_regression_gradient(double_matrix_t *theta, double_matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y, double lambda) {
return logistic_regression_gradient_params(theta, gradient, x, y, p_y, NULL, lambda);
}