void digestBuf()
{
// Mangle the data
for(size_t j = 0; j < 3; j++)
{
// Taint the bytes
GAssert(DIGEST_BYTES >= 32);
if(m_pTaint)
Taint_buffer(m_pBuf, m_pTaint + (m_rand.next() % (DIGEST_BYTES - 32)), 32, j);
// Shuffle the bytes
shuffleBytes(m_pBuf, DIGEST_BYTES);
// Hash each 64-bit chunk with sha-512
unsigned char* pChunk = m_pBuf;
for(size_t i = 0; i < (DIGEST_BYTES / 64); i++)
{
sha512_ctx ctx;
sha512_begin(&ctx);
sha512_hash(pChunk, 64, &ctx);
sha512_end(pChunk, &ctx);
pChunk += 64;
}
}
// Add it to the hash
uint64_t* pA = (uint64_t*)m_pBuf;
uint64_t* pB = (uint64_t*)m_pHash;
for(size_t i = 0; i < (DIGEST_BYTES / sizeof(uint64_t)); i++)
{
(*pB) += (*pA);
pB++;
pA++;
}
}
void Iterate(bool forw)
{
g_updateView = m_pUpdateDisplay->isChecked();
if(!g_updateView && m_prng->next(8192) == 0)
g_updateView = true;
m_pModel->IterateModel(m_pBullets->selection(), forw);
m_pCanvas->setImage(m_pImage);
}
void wipe(unsigned char* pBuf, size_t len)
{
while(len > 0)
{
*(pBuf++) = (unsigned char)m_rand.next();
len--;
}
}
void GLinearRegressor_linear_test(GRand& prng)
{
// Train
GMatrix features1(0, 3);
GMatrix labels1(0, 1);
for(size_t i = 0; i < 1000; i++)
{
double* pVec = features1.newRow();
pVec[0] = prng.uniform();
pVec[1] = prng.uniform(); // irrelevant attribute
pVec[2] = prng.uniform();
labels1.newRow()[0] = 0.3 * pVec[0] + 2.0 * pVec[2] + 5.0;
}
GLinearRegressor lr(prng);
lr.train(features1, labels1);
// Check some values
if(lr.beta()->rows() != 1 || lr.beta()->cols() != 3)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[0] - 0.3) > 1e-6)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[1] - 0.0) > 1e-6)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[2] - 2.0) > 1e-6)
throw Ex("failed");
if(std::abs(lr.epsilon()[0] - 5.0) > 1e-6)
throw Ex("failed");
// Test
GMatrix features2(0, 3);
GMatrix labels2(0, 1);
for(size_t i = 0; i < 1000; i++)
{
double* pVec = features2.newRow();
pVec[0] = prng.uniform();
pVec[1] = prng.uniform(); // irrelevant attribute
pVec[2] = prng.uniform();
labels2.newRow()[0] = 0.3 * pVec[0] + 2.0 * pVec[2] + 5.0;
}
double results;
lr.accuracy(features2, labels2, &results);
if(results > 0.0005)
throw Ex("failed");
}
void GLinearRegressor_linear_test(GRand& prng)
{
// Train
GMatrix features1(0, 3);
GMatrix labels1(0, 1);
for(size_t i = 0; i < 1000; i++)
{
GVec& vec = features1.newRow();
vec[0] = prng.uniform();
vec[1] = prng.uniform(); // irrelevant attribute
vec[2] = prng.uniform();
labels1.newRow()[0] = 0.3 * vec[0] + 2.0 * vec[2] + 5.0;
}
GLinearRegressor lr;
lr.train(features1, labels1);
// Check some values
if(lr.beta()->rows() != 1 || lr.beta()->cols() != 3)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[0] - 0.3) > 1e-6)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[1] - 0.0) > 1e-6)
throw Ex("failed");
if(std::abs(lr.beta()->row(0)[2] - 2.0) > 1e-6)
throw Ex("failed");
if(std::abs(lr.epsilon()[0] - 5.0) > 1e-6)
throw Ex("failed");
// Test
GMatrix features2(0, 3);
GMatrix labels2(0, 1);
for(size_t i = 0; i < 1000; i++)
{
GVec& vec = features2.newRow();
vec[0] = prng.uniform();
vec[1] = prng.uniform(); // irrelevant attribute
vec[2] = prng.uniform();
labels2.newRow()[0] = 0.3 * vec[0] + 2.0 * vec[2] + 5.0;
}
double rmse = sqrt(lr.sumSquaredError(features2, labels2) / features2.rows());
if(rmse > 0.0224)
throw Ex("failed");
}
unsigned char* getHash()
{
if(m_remaining < DIGEST_BYTES)
{
while(m_remaining > 0)
{
*m_pPos = (unsigned char)m_rand.next();
m_pPos++;
m_remaining--;
}
digestBuf();
m_remaining = DIGEST_BYTES;
}
return m_pHash;
}
virtual void doJob(size_t jobId)
{
// Randomly draw some data (with replacement)
GReleaseDataHolder hDrawnFeatures(&m_drawnFeatures);
GReleaseDataHolder hDrawnLabels(&m_drawnLabels);
for(size_t j = 0; j < m_drawSize; j++)
{
size_t r = (size_t)m_rand.next(m_features.rows());
m_drawnFeatures.takeRow((GVec*)&m_features[r]); // This case is only okay because we only use drawFeatures as a const GMatrix
m_drawnLabels.takeRow((GVec*)&m_labels[r]); // This case is only okay because we only use drawnLabels as a const GMatrix
}
// Train the learner with the drawn data
m_pBag->models()[jobId]->m_pModel->train(m_drawnFeatures, m_drawnLabels);
}
void update()
{
m_x = MAX(0, MIN(200, m_x + (int)m_rand.next(15) - 7));
m_y = MAX(0, MIN(500, m_y + (int)m_rand.next(15) - 7));
}
void resetSeed()
{
m_rand.setSeed(0);
}
void shuffleBytes(unsigned char* pBuf, size_t len)
{
for(size_t i = len; i > 1; i--)
std::swap(pBuf[(size_t)m_rand.next(i)], pBuf[i - 1]);
}
