-
Notifications
You must be signed in to change notification settings - Fork 0
/
lcsgd.cc
84 lines (66 loc) · 2.08 KB
/
lcsgd.cc
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
#include "lcsgd.h"
#include <cmath>
#include <iostream>
namespace lcsgd {
LCSGD::LCSGD() : updateN_(0), alpha_(20.0), lambda_(0.00001) {}
LCSGD::~LCSGD() {}
void LCSGD::LoadData(const data_t& data) {
data_ = data;
Initialize();
}
void LCSGD::Initialize() {
std::vector<double>(data_.data_size).swap(latest_coeffs_);
weight_ = dense_vector_t::Zero(data_.max_feature_id + 1);
average_subgradient_ = dense_vector_t::Zero(data_.max_feature_id + 1);
}
int LCSGD::Update(int iterN) {
for (int iter = 0; iter < iterN; ++iter) {
UpdateOnce();
++updateN_;
}
return 0;
}
void LCSGD::UpdateOnce() {
std::uniform_int_distribution<int> dist(1, data_.data_size);
int index = dist(engine_) - 1;
UpdateAverageSubgradient(index);
weight_ = (1.0 - alpha_ * lambda_) * weight_
- alpha_ / data_.data_size * average_subgradient_;
}
void LCSGD::UpdateAverageSubgradient(int index) {
double prev_coeff = latest_coeffs_[index];
double current_coeff = CalcNewCoeff(index);
latest_coeffs_[index] = current_coeff;
AddSubgradient2AS(index, current_coeff - prev_coeff);
}
void LCSGD::AddSubgradient2AS(int index, double coeff) {
const features_t& fv = data_.examples[index].features;
for (auto it = fv.begin(); it != fv.end(); ++it) {
average_subgradient_(it->first) += coeff * it->second;
}
}
double LCSGD::CalcNewCoeff(int index) {
const datum_t& datum = data_.examples[index];
double score = CalcScore(datum);
double coeff = - datum.label * (1.0 / (std::exp(score) + 1.0));
return coeff;
}
double LCSGD::CalcScore(const datum_t& datum) {
const features_t& features = datum.features;
const binary_label_t& label = datum.label;
double score = 0.0;
for (auto it = features.begin(); it != features.end(); ++it) {
score += weight_(it->first) * it->second;
}
return label * score;
}
double LCSGD::Evaluation() {
double result = 0.0;
for (auto it = data_.examples.begin(); it != data_.examples.end(); ++it) {
double score = CalcScore(*it);
result += std::log(1.0 + std::exp(-score));
}
result += weight_.norm();
return result;
}
} //namespace lcsgd