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
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
|
/*
Copyright (c) by respective owners including Yahoo!, Microsoft, and
individual contributors. All rights reserved. Released under a BSD (revised)
license as described in the file LICENSE.
*/
#include <stdint.h>
#include "parse_primitives.h"
#include "v_array.h"
#include "example.h"
#include "simple_label.h"
#include "gd.h"
#include "global_data.h"
int compare_feature(const void* p1, const void* p2) {
feature* f1 = (feature*) p1;
feature* f2 = (feature*) p2;
return (f1->weight_index - f2->weight_index);
}
float collision_cleanup(v_array<feature>& feature_map) {
int pos = 0;
float sum_sq = 0.;
for(uint32_t i = 1;i < feature_map.size();i++) {
if(feature_map[i].weight_index == feature_map[pos].weight_index)
feature_map[pos].x += feature_map[i].x;
else {
sum_sq += feature_map[pos].x*feature_map[pos].x;
feature_map[++pos] = feature_map[i];
}
}
sum_sq += feature_map[pos].x*feature_map[pos].x;
feature_map.end = &(feature_map[pos]);
feature_map.end++;
return sum_sq;
}
struct features_and_source
{
v_array<feature> feature_map; //map to store sparse feature vectors
weight* base;
};
void vec_store(features_and_source& p, float fx, float& fw) {
feature f = {fx, (uint32_t)(&fw - p.base)};
p.feature_map.push_back(f);
}
namespace VW {
flat_example* flatten_example(vw& all, example *ec)
{
flat_example* fec = (flat_example*) calloc(1,sizeof(flat_example));
fec->ld = ec->ld;
fec->final_prediction = ec->final_prediction;
fec->tag_len = ec->tag.size();
if (fec->tag_len >0)
{
fec->tag = ec->tag.begin;
}
fec->example_counter = ec->example_counter;
fec->ft_offset = ec->ft_offset;
fec->global_weight = ec->global_weight;
fec->num_features = ec->num_features;
features_and_source fs;
fs.base = all.reg.weight_vector;
GD::foreach_feature<features_and_source, vec_store>(all, *ec, fs);
qsort(fs.feature_map.begin, fs.feature_map.size(), sizeof(feature), compare_feature);
fec->feature_map_len = fs.feature_map.size();
if (fec->feature_map_len > 0)
{
fec->feature_map = fs.feature_map.begin;
}
return fec;
}
void free_flatten_example(flat_example* fec)
{
if (fec)
free(fec);
}
}
example *alloc_examples(size_t label_size, size_t count=1)
{
example* ec = (example*)calloc(count, sizeof(example));
if (ec == NULL) return NULL;
for (size_t i=0; i<count; i++) {
ec[i].ld = calloc(1, label_size);
if (ec[i].ld == NULL) {
for (size_t j=0; j<i; j++) free(ec[j].ld);
free(ec);
return NULL;
}
ec[i].in_use = true;
ec[i].ft_offset = 0;
// std::cerr << " alloc_example.indices.begin=" << ec->indices.begin << " end=" << ec->indices.end << " // ld = " << ec->ld << "\t|| me = " << ec << std::endl;
}
return ec;
}
void dealloc_example(void(*delete_label)(void*), example&ec)
{
if (delete_label) {
delete_label(ec.ld);
}
ec.tag.delete_v();
ec.topic_predictions.delete_v();
free(ec.ld);
for (size_t j = 0; j < 256; j++)
{
ec.atomics[j].delete_v();
if (ec.audit_features[j].begin != ec.audit_features[j].end_array)
{
for (audit_data* temp = ec.audit_features[j].begin;
temp != ec.audit_features[j].end; temp++)
if (temp->alloced) {
free(temp->space);
free(temp->feature);
temp->alloced = false;
}
ec.audit_features[j].delete_v();
}
}
ec.indices.delete_v();
}
audit_data copy_audit_data(audit_data &src) {
audit_data dst;
dst.space = (char*)calloc(strlen(src.space)+1, sizeof(char));
strcpy(dst.space, src.space);
dst.feature = (char*)calloc(strlen(src.feature)+1, sizeof(char));
strcpy(dst.feature, src.feature);
dst.weight_index = src.weight_index;
dst.x = src.x;
dst.alloced = src.alloced;
return dst;
}
namespace VW {
void copy_example_label(example*dst, example*src, size_t label_size, void(*copy_label)(void*&,void*)) {
if (!src->ld) {
if (dst->ld) free(dst->ld); // TODO: this should be a delete_label, really
dst->ld = NULL;
} else {
if ((label_size == 0) && (copy_label == NULL)) {
if (dst->ld) free(dst->ld); // TODO: this should be a delete_label, really
dst->ld = NULL;
} else if (copy_label) {
copy_label(dst->ld, src->ld);
} else {
//dst->ld = (void*)malloc(label_size);
memcpy(dst->ld, src->ld, label_size);
}
}
}
void copy_example_data(bool audit, example* dst, example* src)
{
//std::cerr << "copy_example_data dst = " << dst << std::endl;
dst->final_prediction = src->final_prediction;
copy_array(dst->tag, src->tag);
dst->example_counter = src->example_counter;
copy_array(dst->indices, src->indices);
for (size_t i=0; i<256; i++)
copy_array(dst->atomics[i], src->atomics[i]);
dst->ft_offset = src->ft_offset;
if (audit)
for (size_t i=0; i<256; i++)
copy_array(dst->audit_features[i], src->audit_features[i], copy_audit_data);
dst->num_features = src->num_features;
dst->partial_prediction = src->partial_prediction;
copy_array(dst->topic_predictions, src->topic_predictions);
dst->loss = src->loss;
dst->eta_round = src->eta_round;
dst->eta_global = src->eta_global;
dst->global_weight = src->global_weight;
dst->example_t = src->example_t;
memcpy(dst->sum_feat_sq, src->sum_feat_sq, 256 * sizeof(float));
dst->total_sum_feat_sq = src->total_sum_feat_sq;
dst->revert_weight = src->revert_weight;
dst->test_only = src->test_only;
dst->end_pass = src->end_pass;
dst->sorted = src->sorted;
dst->in_use = src->in_use;}
void copy_example_data(bool audit, example* dst, example* src, size_t label_size, void(*copy_label)(void*&,void*)) {
copy_example_data(audit, dst, src);
copy_example_label(dst, src, label_size, copy_label);
}
}
|
