-
Notifications
You must be signed in to change notification settings - Fork 0
/
runtests.cpp
455 lines (395 loc) · 11.8 KB
/
runtests.cpp
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
#include <fstream>
#include <iostream>
#include <memory>
#include <Eigen/Dense>
#include "datatypes.hpp"
#include "utility.hpp"
std::default_random_engine generator;
template<typename T>
void
writeCsvMatrix (std::ostream &fd,
const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> &data,
const Eigen::VectorXi &alloc)
{
fd << "cluster";
for (int i = 0; i < data.cols(); i++) {
fd << ",f" << i+1;
}
fd << "\n";
for (int i = 0; i < data.rows(); i++) {
fd << "c" << alloc(i)+1;
for (int j = 0; j < data.cols(); j++) {
fd << "," << data(i,j);
}
fd << "\n";
}
}
template<typename T> void
checkEqualTemplate (T a, T b, const std::string &msg)
{
if (a == b)
return;
std::cerr
<< "Error: " << msg
<< " (expecting " << a << " == " << b << ")"
<< std::endl;
}
void checkEqual(int a, int b, const std::string &msg) {
return checkEqualTemplate(a, b, msg); }
template<typename T> void
checkApproxEqual(const T &a, const T &b,
const std::string &msg,
typename T::Scalar epsilon = 1e-6)
{
// Eigen::Array<bool, T::RowsAtCompileTime, T::ColsAtCompileTime>
// em = (a - b).array() <= a.array().abs().min(b.array().abs()) * epsilon;
if (a.isApprox(b, epsilon))
return;
std::cerr
<< "Error: " << msg
<< " (matrix equality check failed at epsilon=" << epsilon << ")"
<< " differences observed: " << std::endl
<< (a-b).array().abs() << std::endl
<< std::endl;
}
void
checkApproxEqual(const double a, const double b,
const std::string &msg, double epsilon = 1e-6)
{
using std::abs;
using std::min;
epsilon *= min(abs(a),abs(b));
if(abs(a-b) <= epsilon)
return;
std::cerr
<< "Error: " << msg
<< " (expecting ||" << a << " - " << b << "|| <= " << epsilon << ")"
<< std::endl;
}
template<typename T> void
checkProbItemMassGt(const Eigen::Matrix<T, Eigen::Dynamic, Eigen::Dynamic> &prob,
int item, T mass, const std::string &msg)
{
// normalise, nans and infinities will propagate!
auto probnorm = prob.array() / prob.sum();
// check all positive
if ((probnorm < 0).any()) {
std::cerr
<< "Error: " << msg
<< " (some probabilities are not positive)\n";
}
if (probnorm(item) < mass) {
std::cerr
<< "Error: " << msg
<< " (mass of item " << item
<< " is " << probnorm(item)
<< ", i.e. < " << mass << ")\n";
}
}
double
nuConditionalAlphaOracle(const Eigen::MatrixXi &alloc)
{
return alloc.rows();
}
double
phiConditionalAlphaOracle(const Eigen::MatrixXi &alloc,
int m, int p)
{
int sum = 0;
for (int i = 0; i < alloc.rows(); i++) {
sum += alloc(i,m) == alloc(i,p);
}
return 1 + sum;
}
double
gammaConditionalAlphaOracle(const Eigen::MatrixXi &alloc,
int m, int jm)
{
int sum = 0;
for (int i = 0; i < alloc.rows(); i++) {
sum += alloc(i,m) == jm;
}
return 1 + sum;
}
/* the following set of functions work by having an @outer function
* that works "down" through the files, setting $j_i$ to the
* appropriate value and then calling the next @outer function. Once
* $j$ has been set for all $K$ files (i.e. @{i == K}), @outer defers
* to @inner and the product for the given $j$ is evaluated.
*/
// See Equation 2 in section "A.2 Normalising Constant" of Kirk et.al 2012
double
nuConditionalBetaOracle(const mdisampler &mdi)
{
const int N = mdi.nclus(), K = mdi.nfiles();
std::vector<int> j(mdi.nfiles());
auto inner = [K,&mdi,&j]() {
long double prod = 1;
for (int k = 0; k < K; k++) {
prod *= mdi.weight(j[k], k);
}
for (int k = 0; k < K-1; k++) {
for (int l = k+1; l < K; l++) {
prod *= 1 + mdi.phi(k, l) * (j[k] == j[l]);
}
}
return prod;
};
std::function<long double(int)> outer = [&outer, &inner, &j, N, K](int i) {
if (i == K)
return inner();
long double sum = 0;
for (j[i] = 0; j[i] < N; j[i]++)
sum += outer(i+1);
return sum;
};
return outer(0);
}
// See Kirk et.al 2012 $b_\phi$ from "Conditional for $\phi_{mp}$"
double
phiConditionalBetaOracle(const mdisampler &mdi,
int m, int p)
{
const int N = mdi.nclus(), K = mdi.nfiles();
std::vector<int> j(mdi.nfiles());
auto inner = [K,&mdi,&j,m,p]() {
long double prod = 1;
for (int k = 0; k < K; k++) {
prod *= mdi.weight(j[k], k);
}
for (int k = 0; k < K-1; k++) {
for (int l = k+1; l < K; l++) {
if (l != p)
prod *= 1 + mdi.phi(k, l) * (j[k] == j[l]);
}
}
for (int k = 0; k < p; k++) {
if (k != m)
prod *= 1 + mdi.phi(k, p) * (j[k] == j[p]);
}
return prod;
};
std::function<long double(int)> outer = [&outer,&inner,&j,N,K,m,p](int i) {
if (i == K)
return inner();
// $j_m$ and $j_p$ are set outside this iteration
if (i == m || i == p)
return outer(i+1);
long double sum = 0;
for (j[i] = 0; j[i] < N; j[i]++)
sum += outer(i+1);
return sum;
};
// iterate over $\sum_{j_m = j_p = 1}^N$
long double sum = 0;
for (int i = 0; i < N; i++) {
j[m] = j[p] = i;
sum += outer(0);
}
return mdi.nu() * sum;
}
// See Kirk et.al 2012 $b_\gamma$ from "Conditional for $\gamma_{j_m m}$"
double
gammaConditionalBetaOracle (const mdisampler &mdi,
const int m, const int jm)
{
const int N = mdi.nclus(), K = mdi.nfiles();
std::vector<int> j(mdi.nfiles());
// within data @m we are interested in cluster @jm
j[m] = jm;
auto inner = [K,m,&mdi,&j]() {
long double prod = 1;
for (int k = 0; k < K; k++) {
if(k != m)
prod *= mdi.weight(j[k], k);
}
for (int k = 0; k < K-1; k++) {
for (int l = k+1; l < K; l++) {
prod *= 1 + mdi.phi(k, l) * (j[k] == j[l]);
}
}
return prod;
};
std::function<long double(int)> outer = [&outer, &inner, &j, N, K, m](int i) {
if (i == K)
return inner();
if (i == m)
return outer(i+1);
long double sum = 0;
for (j[i] = 0; j[i] < N; j[i]++)
sum += outer(i+1);
return sum;
};
return mdi.nu() * outer(0);
}
int
main()
{
/* datatypes
*
* * loading data (CPU)
*
* * summarising data given cluster allocations (CPU GPU)
*
* * sampling cluster parameters given above summary (CPU GPU)
*
* * sampling cluster allocations given cluster parameters (CPU GPU)
*
* MDI prior
*
* * Sampling Nu given [weights and allocations]?
*
* * Sampling weights given [lots!]
*
* * Sampling DPs given weights and nu
*/
generator.seed(1);
// given a single "file" for each datatype, load it in define
// allocations (same for all four files), weights, nu, DP
// concentration
const int
nfiles = 4,
nitems = 5,
nclus = 10,
ngaussfeatures = 3;
// 5 items, cluster allocations [0 0 1 1 2]. one file for each
// datatype
Eigen::MatrixXi alloc(nitems, nfiles);
for (int i = 0; i < nfiles; i++)
alloc.col(i) << 0, 0, 1, 1, 2;
Eigen::MatrixXf
weights = alloc.cast<float>();
Eigen::VectorXf
dpmass = Eigen::VectorXf::Ones(nfiles);
Eigen::MatrixXf data_gaussian(nitems, ngaussfeatures);
data_gaussian <<
-2.1, -2.1, -2.1,
-1.9, -1.9, -1.9,
1.9, 1.9, 1.9,
2.1, 2.1, 2.1,
5.0, 5.0, 5.0;
// write dummy data out, load it back in and make sure it's the same
// as the original data
{
std::ofstream out("data_gauss.txt");
writeCsvMatrix (out, data_gaussian, alloc.col(0));
}
gaussianDatatype dt_gauss("data_gauss.txt");
checkEqual(nitems, dt_gauss.items().size(),
"number of items read from gaussian dataset");
checkEqual(ngaussfeatures, dt_gauss.features().size(),
"number of features read from gaussian dataset");
checkApproxEqual<Eigen::MatrixXf>(data_gaussian.transpose(), dt_gauss.rawdata(),
"gaussian data from file");
// given weights, allocations, nu:
shared shared(nfiles, nclus, nitems);
interdataset inter(nfiles);
mdisampler mdi(inter, nclus);
shared.sampleFromPrior();
mdi.sampleFromPrior();
shared.setAlloc(alloc);
cuda::sampler cuda(nfiles, nitems, nclus,
inter.getPhiOrd(), inter.getWeightOrd());
cuda.setNu(mdi.nu());
cuda.setDpMass(eigenMatrixToStdVector(mdi.dpmass()));
cuda.setAlloc(eigenMatrixToStdVector(alloc));
cuda.setPhis(eigenMatrixToStdVector(mdi.phis()));
cuda.setWeights(eigenMatrixToStdVector(mdi.weights()));
gaussianSampler * gauss_sampler = dt_gauss.newSampler(nclus, &cuda);
// given known allocations, calculate Gaussian summaries
{
gauss_sampler->cudaSampleParameters(alloc.col(0));
gauss_sampler->cudaAccumAllocProbs();
const std::vector<runningstats<> > state(gauss_sampler->accumState(alloc.col(0)));
checkEqual(nclus * ngaussfeatures, state.size(),
"number of accumulated gaussian stats");
// check Gaussian summaries are OK
const runningstats<> rs[nclus] = {{2,-2,0.02},{2,2,0.02},{1,5,0}};
bool ok = true;
for (int j = 0; j < nclus; j++) {
for (int i = 0; i < ngaussfeatures; i++) {
const runningstats<>
&a = state[j * ngaussfeatures + i],
&b = rs[j];
ok &= a.isApprox(b);
}
}
if (!ok) {
std::cout << "Error: some of the accumulated gaussian stats are incorrect\n";
}
}
// given known cluster parameters, ...
{
Eigen::MatrixXf mu(ngaussfeatures, nclus), tau(ngaussfeatures,nclus);
mu.fill(0); tau.fill(20);
mu.col(0).fill(-2);
mu.col(1).fill( 2);
mu.col(2).fill( 5);
gauss_sampler->debug_setMuTau(mu, tau);
}
// ... sample Gaussian cluster association probabilities
{
std::unique_ptr<sampler::item> is(gauss_sampler->newItemSampler());
for (int i = 0; i < nitems; i++) {
Eigen::VectorXf prob((*is)(i));
prob = (prob.array() - prob.maxCoeff()).exp();
checkProbItemMassGt<float>(prob, alloc(i,0), 0.5,
"gaussian cluster allocations");
}
}
{
double
oracle = nuConditionalBetaOracle(mdi);
checkApproxEqual(oracle, mdi.nuConditionalBeta(),
"MDI nu beta conditional, CPU code");
checkApproxEqual(oracle, cuda.collectNuConditionalBeta(),
"MDI nu beta conditional, GPU code");
oracle = nuConditionalAlphaOracle(alloc);
checkApproxEqual(oracle, shared.nuConditionalAlpha(),
"MDI nu alpha conditional, CPU code");
}
{
Eigen::MatrixXf oracle, cpu, gpu;
oracle = cpu = gpu = Eigen::MatrixXf::Zero(nfiles, nfiles);
std::vector<float> gpuv = cuda.collectPhiConditionalsBeta();
for (int k = 0, i = 0; k < nfiles; k++) {
for (int l = k+1; l < nfiles; l++) {
oracle(k,l) = phiConditionalBetaOracle(mdi, k, l);
cpu(k,l) = mdi.phiConditionalBeta(k, l);
gpu(k,l) = gpuv[i++];
}
}
checkApproxEqual(oracle, cpu, "MDI phi beta conditionals, CPU code");
checkApproxEqual(oracle, gpu, "MDI phi beta conditionals, GPU code");
gpuv = cuda.collectPhiConditionalsAlpha();
for (int k = 0, i = 0; k < nfiles; k++) {
for (int l = k+1; l < nfiles; l++) {
oracle(k,l) = phiConditionalAlphaOracle(alloc, k, l);
cpu(k,l) = shared.phiConditionalAlpha(k, l) + 1; // what to do about prior?
gpu(k,l) = gpuv[i++];
}
}
checkApproxEqual(oracle, cpu, "MDI phi alpha conditionals, CPU code");
checkApproxEqual(oracle, gpu, "MDI phi alpha conditionals, GPU code");
}
{
Eigen::MatrixXf oracle, cpu, gpu;
oracle = cpu = Eigen::MatrixXf::Zero(nclus, nfiles);
gpu = stdVectorToEigenMatrix(cuda.collectGammaConditionalsBeta(),
nclus, nfiles);
// take out the prior, not sure what to do about this. I think I
// should be checking priors as well, but conflating it in the
// test unnecessarily seems to make things more difficult to debug
// than it could... hum!
gpu.array() -= 1;
for (int k = 0; k < nfiles; k++) {
for (int j = 0; j < nclus; j++) {
oracle(j,k) = gammaConditionalBetaOracle(mdi, k, j);
cpu(j,k) = mdi.gammaConditionalBeta(k, j);
}
}
checkApproxEqual(oracle, cpu, "MDI gamma beta conditionals, CPU code");
checkApproxEqual(oracle, gpu, "MDI gamma beta conditionals, GPU code");
}
return 0;
}