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

[classification] removing regularization update from gradient computation in logistic regression, as that's now handled by the optimizer

Browse Source
This commit is contained in:
Al
2017-04-02 14:32:14 -04:00
parent 64c049730a
commit a4431dbb27
2 changed files with 52 additions and 107 deletions
Download Patch File Download Diff File Expand all files Collapse all files

152
src/logistic_regression.c
View File

@@ -4,12 +4,16 @@
#include "logistic.h"
#include "file_utils.h"
#define NEAR_ZERO_WEIGHT 1e-6
bool logistic_regression_model_expectation_sparse(sparse_matrix_t *theta, sparse_matrix_t *x, double_matrix_t *p_y) {
if (theta == NULL || x == NULL || p_y == NULL) {
if (theta == NULL) log_error("theta = NULL\n");
if (x == NULL) log_error("x = NULL\n");
if (p_y == NULL) log_error("p_y = NULL\n");
return false;
}
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) {
if (sparse_matrix_dot_sparse(x, theta, p_y) != 0) {
log_error("x->m = %zu, x->n = %zu, theta->m = %zu, theta->n = %zu, p_y->m = %zu, p_y->n = %zu\n", x->m, x->n, theta->m, theta->n, p_y->m, p_y->n);
return false;
}
@@ -18,16 +22,41 @@ bool logistic_regression_model_expectation(double_matrix_t *theta, sparse_matrix
return true;
}
double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y, double lambda) {
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) {
if (theta == NULL) log_error("theta = NULL\n");
if (x == NULL) log_error("x = NULL\n");
if (p_y == NULL) log_error("p_y = NULL\n");
return false;
}
if (sparse_matrix_dot_dense(x, theta, p_y) != 0) {
log_error("x->m = %zu, x->n = %zu, theta->m = %zu, theta->n = %zu, p_y->m = %zu, p_y->n = %zu\n", x->m, x->n, theta->m, theta->n, p_y->m, p_y->n);
return false;
}
softmax_matrix(p_y);
return true;
}
double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y) {
size_t m = x->m;
size_t n = x->n;
if (m != y->n) return -1.0;
if (!double_matrix_resize(p_y, x->m, theta->n)) {
if (m != y->n) {
log_error("m = %zu, y->n = %zu\n", m, y->n);
return -1.0;
}
if (!double_matrix_resize_aligned(p_y, x->m, theta->n, 16)) {
log_error("resize_aligned failed\n");
return -1.0;
}
double_matrix_zero(p_y);
if (!logistic_regression_model_expectation(theta, x, p_y)) {
log_error("model expectation failed\n");
return -1.0;
}
@@ -37,39 +66,23 @@ double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t
for (size_t i = 0; i < p_y->m; i++) {
uint32_t y_i = y->a[i];
double value = double_matrix_get(p_y, i, y_i);
if (value > NEAR_ZERO_WEIGHT) {
cost += log(value);
}
cost += log(value);
}
cost *= -(1.0 / m);
if (lambda > 0.0) {
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 = double_matrix_get(theta, i, j);
reg_cost += theta_ij * theta_ij;
}
}
cost += reg_cost * (lambda / 2.0);
}
return cost;
}
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) {
bool logistic_regression_gradient(double_matrix_t *theta, double_matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y) {
size_t m = x->m;
size_t n = x->n;
if (m != y->n) return false;
if (m != y->n || theta->m != gradient->m || theta->n != gradient->n) return false;
if (!double_matrix_resize(p_y, x->m, theta->n) || !double_matrix_resize(p_y, x->m, theta->n)) {
if (!double_matrix_resize_aligned(p_y, x->m, theta->n, 16)) {
return false;
}
double_matrix_zero(p_y);
if (!logistic_regression_model_expectation(theta, x, p_y)) {
return false;
@@ -85,93 +98,26 @@ static bool logistic_regression_gradient_params(double_matrix_t *theta, double_m
uint32_t col;
double data;
bool regularize = lambda > 0.0;
double_matrix_zero(gradient);
double *theta_values = theta->values;
double *predicted_values = p_y->values;
double *gradient_values = gradient->values;
// Zero the relevant rows of the gradient
if (x_cols != NULL) {
double *gradient_i;
size_t batch_rows = x_cols->n;
uint32_t *cols = x_cols->a;
for (i = 0; i < batch_rows; i++) {
col = cols[i];
gradient_i = double_matrix_get_row(gradient, col);
double_array_zero(gradient_i, num_classes);
}
} else {
double_matrix_zero(gradient);
}
// gradient = -(1. / m) * x.T.dot(y - p_y) + lambda * theta
// gradient = -(1. / m) * x.T.dot(y - p_y)
sparse_matrix_foreach(x, row, col, data, {
uint32_t y_i = y->a[row];
for (j = 0; j < num_classes; j++) {
double class_prob = predicted_values[row * num_classes + j];
double class_prob = double_matrix_get(p_y, row, j);
double residual = (y_i == j ? 1.0 : 0.0) - class_prob;
gradient_values[col * num_classes + j] += data * residual;
double gradient_update = data * residual;
double_matrix_add_scalar(gradient, col, j, gradient_update);
}
})
double scale = -1.0 / m;
// Scale the vector by -1 / m using only the unique columns in X
// Useful for stochastic and minibatch gradients
if (x_cols != NULL) {
size_t batch_rows = x_cols->n;
uint32_t *cols = x_cols->a;
for (i = 0; i < batch_rows; i++) {
col = cols[i];
for (j = 0; j < num_classes; j++) {
gradient_values[col * num_classes + j] *= scale;
}
}
} else {
double_matrix_mul(gradient, scale);
}
// Update the only the relevant columns in x
if (regularize) {
size_t num_rows = num_features;
uint32_t *cols = NULL;
if (x_cols != NULL) {
cols = x_cols->a;
num_rows = x_cols->n;
}
for (i = 0; i < num_rows; i++) {
col = x_cols != NULL ? cols[i] : i;
for (j = 0; j < num_classes; j++) {
size_t idx = col * num_classes + j;
double theta_ij = theta_values[idx];
double reg_update = theta_ij * lambda;
double current_value = gradient_values[idx];
double updated_value = current_value + reg_update;
if ((updated_value > 0) == (current_value > 0)) {
gradient_values[idx] = updated_value;
}
}
}
}
double_matrix_mul(gradient, scale);
return true;
}
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(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);
}

7
src/logistic_regression.h
View File

@@ -27,9 +27,8 @@ may be called softmax regression.
#include "sparse_matrix.h"
bool logistic_regression_model_expectation(double_matrix_t *theta, sparse_matrix_t *x, double_matrix_t *p_y);
double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y, double lambda);
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);
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);
bool logistic_regression_model_expectation_sparse(sparse_matrix_t *theta, sparse_matrix_t *x, double_matrix_t *p_y);
double logistic_regression_cost_function(double_matrix_t *theta, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y);
bool logistic_regression_gradient(double_matrix_t *theta, double_matrix_t *gradient, sparse_matrix_t *x, uint32_array *y, double_matrix_t *p_y);
#endif