-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathSGD.cpp
More file actions
535 lines (471 loc) · 16.3 KB
/
Copy pathSGD.cpp
File metadata and controls
535 lines (471 loc) · 16.3 KB
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
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
#include <iostream>
#include <fstream>
#include <vector>
#include <string>
#include <sstream>
#include <float.h>
#include <cmath>
#include <numeric>
#include <random>
#include <thread>
using namespace std;
typedef long double ld;
// (X*W) + B
vector<ld> predict(vector<vector<ld>>X, vector<ld>pW, ld pB){
int batch_size = X.size();
int features = X[0].size();
vector<ld> predicted(batch_size, 0.0);
for(int i=0;i<batch_size;i++){
for(int j=0;j<features;j++){
predicted[i] += X[i][j]*pW[j];
}
predicted[i] += pB;
}
return predicted;
}
// for l1 regularization
vector<ld> predict_minus_j(vector<vector<ld>> X, vector<ld> W, ld B, int j){
int batch_size = X.size();
int features = X[0].size();
vector<ld> predicted(batch_size, 0.0);
for(int i=0;i<batch_size;i++){
for(int k=0;k<features;k++){
if(k == j){
predicted[i] += 0;
}
else{
predicted[i] += X[i][k]*W[k];
}
}
predicted[i] += B;
}
return predicted;
}
// store mean of all features in a vector
vector<ld> mean_vector (vector<vector<ld>> dataset){
int featureset_row = dataset.size();
int featureset_col = dataset[0].size() - 1;
vector<ld> mean(featureset_col, 0.0);
for(int i=0;i<featureset_col;i++){
ld sum = 0.0;
for(int j=0;j<featureset_row;j++){
sum += dataset[j][i];
}
mean[i] = sum/featureset_row;
}
return mean;
}
// store standard deviation of all features in a vector
vector<ld> standard_deviation_vector(vector<vector<ld>> dataset, vector<ld> mean){
int featureset_row = dataset.size();
int featureset_col = dataset[0].size() - 1;
vector<ld> sd(featureset_col, 0.0);
for(int i=0;i<featureset_col;i++){
ld sum = 0.0;
for(int j=0;j<featureset_row;j++){
sum += pow((dataset[j][i]-mean[i]), 2);
}
sd[i] = pow((sum/featureset_row), 0.5);
}
return sd;
}
// vector A = vector A - vector B
vector<ld> vector_subtraction(vector<ld> a, vector<ld> b){
for(int i=0;i<a.size();i++){
a[i] -= b[i];
}
return a;
}
// vector A = vector A - vector B
vector<ld> vector_scaler_addition(ld a, vector<ld>b){
for(int i=0;i<b.size();i++){
b[i] += a;
}
return b;
}
// vector C = vector A x scaler B
vector<ld> vector_scaler_multiplication(ld value, vector<ld> X){
for(int i=0;i<X.size();i++){
X[i] *= value;
}
return X;
}
// transpose of a vector(matrix)
vector<vector<ld>> transpose_vector(vector<vector<ld>> X){
int batch_size = X.size();
int features = X[0].size();
vector<vector<ld>> transpose(features, vector<ld>(batch_size));
for(int i=0;i<batch_size;i++){
for(int j=0;j<features;j++){
transpose[j][i] = X[i][j];
}
}
return transpose;
}
// vector multiplication, (N x M)*(M x 1) = N x 1
vector<ld> vector_multiplication(vector<ld> y, vector<vector<ld>> XT){
int features = XT.size();
int batch_size = XT[0].size();
vector<ld> result(features, 0.0);
for(int i=0;i<features;i++){
for(int j=0;j<batch_size;j++){
result[i] += XT[i][j]*y[j];
}
}
return result;
}
// mean square error
ld MSE(vector<ld> y, vector<ld> predicted){
int size = y.size();
ld mse = 0.0;
for(int i=0;i<size;i++){
mse += pow((y[i] - predicted[i]), 2);
}
return mse/size;
}
// mean absolute error
ld MAE(vector<ld> y, vector<ld> predicted){
int size = y.size();
ld mae = 0.0;
for(int i=0;i<size;i++){
mae += abs(y[i] - predicted[i]);
}
return mae/size;
}
// Rsquared error
ld Rsquared(vector<ld> y, vector<ld> predicted){
int size = y.size();
ld sse = 0.0;
ld sst = 0.0;
ld y_mean = (accumulate(y.begin(), y.end(), (ld)0.0))/size;
for(int i=0;i<size;i++){
sse += pow((y[i] - predicted[i]), 2);
sst += pow((y[i] - y_mean), 2);
}
return (1.0 - (sse/sst));
}
// initialize vector with random values
void initialize(vector<ld> &v) {
srand (static_cast <unsigned> (time(nullptr)));
for (int i = 0; i < v.size(); i++) {
ld r = static_cast<float>(rand())/static_cast<ld>(RAND_MAX);
v[i] = r;
}
}
// scaling the dataset using Mean Normalization technique
vector<vector<ld>> scale(vector<vector<ld>> dataset, vector<ld> mean, vector<ld> sd){
int featureset_row = dataset.size();
int featureset_col = dataset[0].size() - 1;
for(int i=0;i<featureset_col;i++){
for(int j=0;j<featureset_row;j++){
dataset[j][i] = (dataset[j][i] - mean[i])/sd[i];
}
}
return dataset;
}
// SGD
void SGD(vector<vector<ld>> training_data, vector<vector<ld>> validation_data, const string& evaluation_metric = "mse", int batch_size = 100, const string& learning_rate = "adaptive", int max_iterations = 10000, ld eta0 = 0.1, bool early_stopping = true, int no_iter_change = 10, bool use_validation_error = true, ld tol = 0.001, bool logging = false, const string& regularization = "none", ld lambda = 0.5){
// for logging to a file
ofstream myfile;
if(logging){
myfile.open("/Users/bis/CLionProjects/test/output.log", ofstream::app);
time_t now = time(nullptr);
char* dt = ctime(&now);
myfile<<"LOCAL DATE AND TIME: "<<dt<<"\n";
myfile<<"Parameters:\n\tbatch_size = "<<batch_size<<"\n\tEvaluation_metric = "<<evaluation_metric<<"\n\tlearning_rate = "<<learning_rate<<"\n\tmax_iterations = "<<max_iterations<<"\n\teta0 = "<<eta0<<"\n\tearly_stopping = "<<early_stopping<<"\n\tno_iter_change = "<<no_iter_change<<"\n\tuse_validation_error = "<<use_validation_error<<"\n\ttolerence = "<<tol<<"\n\tregularization = "<<regularization<<"\n\tlambda = "<<lambda<<"\n\n";
}
// generating m = batch_size unique random numbers
int n = training_data.size();
int arr[n];
for(int i=0;i<n;i++){
arr[i] = i;
}
// initializing variables
int features = training_data[0].size() - 1;
// initialize X and y for validation set
vector<vector<ld>> vX(validation_data.size(), vector<ld>(features));
vector<ld> vy(validation_data.size());
// if using validation error
if(use_validation_error){
for(int x=0;x<validation_data.size();x++){
vector<ld> data = validation_data[x];
vy[x] = data.back();
data.pop_back();
vX[x] = data;
}
}
static vector<ld> W(features, 0.0);
// initialize W with random values only for the first time
if(accumulate(W.begin(), W.end(), (ld)0.0) == 0.0){
initialize(W);
}
static ld B = 0.0;
vector<ld> stepsize_W(features);
vector<ld> pW(features);
vector<ld> dW(features);
vector<ld> dowW(batch_size, DBL_MAX);
ld pB;
ld dB;
ld pE = INFINITY;
ld error;
for(int i=0;i<max_iterations;i++){
// shuffle arr for random values
auto rng = default_random_engine (random_device()());
shuffle(arr, arr+n, rng);
// initialize X and y
vector<vector<ld>> X(batch_size, vector<ld>(features));
vector<ld> y(batch_size);
for(int x=0;x<batch_size;x++){
// Getting random data point(s)
vector<ld> data = training_data[arr[x]];
// store and remove class label
y[x] = data.back();
data.pop_back();
X[x] = data;
}
if(logging){
myfile<<"*********************** ITERATION "<<i+1<<" ***********************\n\n";
}
cout<<"*********************** ITERATION "<<i+1<<" ***********************\n\n";
// store the weights and bias
pW = W;
pB = B;
// find the derivatives
if(regularization == "none"){ // for no regularization
dowW = vector_subtraction(y, predict(X, pW, pB));
dW = vector_multiplication(dowW, transpose_vector(X));
dW = vector_scaler_multiplication((-2.0/batch_size), dW);
dB = (-2.0/batch_size)*accumulate(dowW.begin(), dowW.end(), (ld)0.0);
// update the weights and bias
W = vector_subtraction(pW, vector_scaler_multiplication(eta0, dW));
B = pB - (eta0*dB);
} else if(regularization == "l2"){ // for l2 regularization
dowW = vector_subtraction(y, predict(X, pW, pB));
dW = vector_multiplication(dowW, transpose_vector(X));
dW = vector_scaler_multiplication((-2.0/batch_size), dW);
ld penalty_weight = 2.0*lambda*accumulate(pW.begin(), pW.end(), (ld)0.0);
dW = vector_scaler_addition(penalty_weight, dW);
dB = (-2.0/batch_size)*accumulate(dowW.begin(), dowW.end(), (ld)0.0);
// update the weights and bias
W = vector_subtraction(pW, vector_scaler_multiplication(eta0, dW));
B = pB - (eta0*dB);
} else if(regularization == "l1"){ // for l1 regularization
for(int w=0;w<W.size();w++){
vector<ld> predicted = predict_minus_j(X, W, pB, w);
ld diff = ((accumulate(y.begin(), y.end(), (ld)0.0)/batch_size) - (accumulate(predicted.begin(), predicted.end(), (ld)0.0)/batch_size));
if(diff < (-lambda/2)){
W[w] = diff + (lambda/2);
} else if(diff > (lambda/2)){
W[w] = diff - (lambda/2);
} else {
W[w] = 0;
}
}
dowW = vector_subtraction(y, predict(X, pW, pB));
dB = (-2.0/batch_size)*accumulate(dowW.begin(), dowW.end(), (ld)0.0);
B = pB - (eta0*dB);
}
// invscaling (inverse scaling) learning rate
if(learning_rate == "invscaling"){
eta0 /= pow(i+1, 0.5);
}
// training error
ld training_error = 0;
if(evaluation_metric == "mae"){
training_error = MAE(y, predict(X, W, B));
}
else if(evaluation_metric == "rsquared"){
training_error = Rsquared(y, predict(X, W, B));
}
else {
training_error = MSE(y, predict(X, W, B));
}
// validation error
ld validation_error = 0;
if(use_validation_error){
if(evaluation_metric == "mae"){
validation_error = MAE(vy, predict(vX, W, B));
}
else if(evaluation_metric == "rsquared"){
validation_error = Rsquared(vy, predict(vX, W, B));
}
else {
validation_error = MSE(vy, predict(vX, W, B));
}
}
if(logging){
myfile<<"Training error: "<<training_error<<"\n";
if(use_validation_error){
myfile<<"Validation error: "<<validation_error<<"\n\n";
}
}
cout<<"Training error: "<<training_error<<"\n\n";
if(use_validation_error){
cout<<"Validation error: "<<validation_error<<"\n\n";
error = validation_error;
} else {
error = training_error;
}
// for Rsquared evaluation metric
if(evaluation_metric == "rsquared"){
error = -error;
}
if(early_stopping){
if(error >= pE-tol && no_iter_change == 0){
if(logging){
myfile<<"Cannot converge more!\n";
myfile<<"Training error: "<<training_error;
if(use_validation_error){
myfile<<"\nValidation error: "<<validation_error<<"\n";
}
}
cout<<"Cannot converge more!\n";
cout<<"Training error: "<<training_error;
if(use_validation_error){
cout<<"\nValidation error: "<<validation_error<<"\n\n";
}
break;
}
else if(error >= pE-tol && no_iter_change != 0){
no_iter_change--;
if(learning_rate == "adaptive"){
eta0 /= 5;
}
}
}
else if(learning_rate == "adaptive" && error >= pE-tol){
eta0 /= 5;
}
pE = error;
}
myfile<<"\n\n~~~~End of Output~~~~\n\n\n";
myfile.close();
}
// preprocess the data and call SGD
void preprocess_and_call_SGD(const vector<vector<ld>>& dataset){
vector<vector<ld>> training_data;
vector<vector<ld>> validation_data;
vector<ld> mean = mean_vector(dataset);
vector<ld> sd = standard_deviation_vector(dataset, mean);
vector<vector<ld>> scaled_dataset = scale(dataset, mean, sd);
int n = scaled_dataset.size();
int arr[n];
for(int i=0;i<n;i++){
arr[i] = i;
}
auto rng = default_random_engine (random_device()());
shuffle(arr, arr+n, rng);
int size = int(0.7*n);
for(int i=0;i<n;i++){
if(i<size){
training_data.push_back(scaled_dataset[arr[i]]);
}
else{
validation_data.push_back(scaled_dataset[arr[i]]);
}
}
SGD(training_data, validation_data, "mse", 1, "adaptive", 1000, 1.0, true, 5, true, 0.001, false, "none", 0.1);
}
// read data from csv file into vector
void read_csv_data(bool buffering = false){
//read data from csv file
vector<vector<ld>> dataset;
int row_count = 0;
string line;
bool ignore_first_column = false;
ifstream dataset_file("/Users/bis/CLionProjects/test/satgpa.csv");
// declare a buffer
int buffer_size = 1;
while(getline(dataset_file, line)){
vector<ld> row;
stringstream ss(line);
string value;
int column_count = 0;
if(row_count == 0){
getline(ss, value, ',');
if(value == "\"\"" || value.empty() || value.length() == 0){
ignore_first_column = true;
}
row_count++;
continue;
}
while(getline(ss, value, ',')){
if(!column_count && ignore_first_column){
column_count++;
continue;
}
//convert string values into ld values
ld val = stold(value);
row.push_back(val);
column_count++;
}
dataset.push_back(row);
if(buffering){
// read the file in chunks of buffer size 1000
if(buffer_size == 1000){
preprocess_and_call_SGD(dataset);
// set buffer_size to 0 after the buffer is full
buffer_size = 0;
// clear the dataset for new batch
dataset.clear();
}
buffer_size++;
}
}
if(!buffering){
preprocess_and_call_SGD(dataset);
}
}
// dummy functions below!
// read data from csv file into vector
void read_csv_data_test(bool buffering = false){
//read data from csv file
vector<vector<string>> dataset;
int row_count = 0;
string line;
bool ignore_first_column = false;
ifstream dataset_file("/Users/bis/Downloads/2017_Yellow_Taxi_Trip_Data.csv");
// declare a buffer
int buffer_size = 1;
int batch_number = 0;
while(getline(dataset_file, line)){
vector<string> row;
stringstream ss(line);
string value;
int column_count = 0;
if(row_count == 0){
getline(ss, value, ',');
if(value == "\"\"" || value.empty() || value.length() == 0){
ignore_first_column = true;
}
row_count++;
continue;
}
while(getline(ss, value, ',')){
if(!column_count && ignore_first_column){
column_count++;
continue;
}
//convert string values into ld values
row.push_back(value);
column_count++;
}
dataset.push_back(row);
if(buffering){
// read the file in chunks of buffer size 1000
if(buffer_size == 1000){
cout<<"Processing batch number "<<++batch_number<<" of "<<dataset.size()<<" rows\n";
// set buffer_size to 0 after the buffer is full
buffer_size = 0;
// clear the dataset for new batch
dataset.clear();
}
buffer_size++;
}
}
cout<<"(last)Processing batch number "<<++batch_number<<" of "<<dataset.size()<<" rows\n";
}
int main(){
read_csv_data(false);
return 0;
}