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

[optimization] for the FTRL and SGD optimizers, use the new *_array_sum_sq function to do L2 regularization, vs. the L2 norm which will use the linear algebra meaning

Browse Source
This commit is contained in:
Al
2018-01-22 01:56:32 -05:00
parent eb3fb37ad4
commit ccce4f793f
3 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
src/stochastic_gradient_descent.c
View File

@@ -295,7 +295,7 @@ double stochastic_gradient_descent_reg_cost(sgd_trainer_t *self, uint32_array *u
double *theta_i = double_matrix_get_row(theta, row);
if (reg_type == REGULARIZATION_L2 && row >= i_start) {
cost += double_array_l2_norm(theta_i, n);
cost += double_array_sum_sq(theta_i, n);
} else if (reg_type == REGULARIZATION_L1 && row >= i_start) {
cost += double_array_l1_norm(theta_i, n);
}