double
distTanimoto(
const MappedColumnVector& inX,
const MappedColumnVector& inY) {
// Note that this is not a metric in general!
double dotProduct = dot(inX, inY);
double tanimoto = inX.squaredNorm() + inY.squaredNorm();
return (tanimoto - 2 * dotProduct) / (tanimoto - dotProduct);
}
AnyType
normalize_sum_array::run(AnyType &args){
const MappedColumnVector input_vector = args[0].getAs<MappedColumnVector>();
const double sum_target = args[1].getAs<double>();
double sum_input_vector = input_vector.sum();
// Avoid divide by zero by dividing by a small number if sum is small
double VAR_IMP_EPSILON = 1e-6;
if (sum_input_vector < VAR_IMP_EPSILON)
sum_input_vector = VAR_IMP_EPSILON;
ColumnVector output_vector = input_vector * sum_target / sum_input_vector;
return output_vector;
}
double
squaredAngle(
const MappedColumnVector& inX,
const MappedColumnVector& inY) {
double cosine = dot(inX, inY) / (inX.norm() * inY.norm());
if (cosine > 1)
cosine = 1;
else if (cosine < -1)
cosine = -1;
double angle = std::acos(cosine);
return angle * angle;
}
AnyType row_fold::run(AnyType & args){
MappedColumnVector vec = args[0].getAs<MappedColumnVector>();
MappedIntegerVector pat = args[1].getAs<MappedIntegerVector>();
if (vec.size() != pat.sum()) {
throw std::invalid_argument(
"dimensions mismatch: row_in.size() != pattern.sum()");
}
ColumnVector r(pat.size());
for (int i = 0, j = 0; i < pat.size(); j += pat[i++])
r[i] = vec.segment(j, pat[i]).prod();
return r;
}
AnyType matrix_vec_mult_in_mem_2d::run(AnyType & args){
MappedColumnVector vec = args[0].getAs<MappedColumnVector>();
MappedMatrix mat = args[1].getAs<MappedMatrix>();
// Note mat is constructed in the column-first order
// which means that mat is actually transposed
if(vec.size() != mat.cols()){
throw std::invalid_argument(
"dimensions mismatch: vec.size() != matrix.rows()");
};
// trans(vec) * trans(mat) = mat * vec
Matrix r = mat * vec;
ColumnVector v = r.col(0);
return v;
}
// -----------------------------------------------------------------------
// Linear regression
// -----------------------------------------------------------------------
AnyType
linregr_transition::run(AnyType& args) {
MutableLinRegrState state = args[0].getAs<MutableByteString>();
if (args[1].isNull() || args[2].isNull()) { return args[0]; }
double y = args[1].getAs<double>();
MappedColumnVector x;
try {
MappedColumnVector xx = args[2].getAs<MappedColumnVector>();
x.rebind(xx.memoryHandle(), xx.size());
} catch (const ArrayWithNullException &e) {
return args[0];
}
state << MutableLinRegrState::tuple_type(x, y);
return state.storage();
}
double
distAngle(
const MappedColumnVector& inX,
const MappedColumnVector& inY) {
// Deal with the undefined case where one of the norm is zero
// Angle is not defined. Just return \pi.
double xnorm = inX.norm(), ynorm = inY.norm();
if (xnorm < std::numeric_limits<double>::denorm_min()
|| ynorm < std::numeric_limits<double>::denorm_min())
return std::acos(-1);
double cosine = dot(inX, inY) / (xnorm * ynorm);
if (cosine > 1)
cosine = 1;
else if (cosine < -1)
cosine = -1;
return std::acos(cosine);
}
AnyType
multi_response_glm_multinom_logit_transition::run(AnyType& args) {
MutableMultiResponseGLMState state = args[0].getAs<MutableByteString>();
if (state.terminated || args[1].isNull() || args[2].isNull()) {
return args[0];
}
double y = args[1].getAs<double>();
MappedColumnVector x;
try {
MappedColumnVector xx = args[2].getAs<MappedColumnVector>();
x.rebind(xx.memoryHandle(), xx.size());
} catch (const ArrayWithNullException &e) {
return args[0];
}
if (state.empty()) {
state.num_features = static_cast<uint16_t>(x.size());
state.num_categories = args[4].getAs<uint16_t>();
state.optimizer.num_coef = static_cast<uint16_t>(
state.num_features * (state.num_categories-1));
// MADLIB-667: GPDB limits the single array size to be 1GB, which means
// that the size of a double array cannot be large than 134217727
// because (134217727 * 8) / (1024 * 1024) = 1023. And solve
// state_size = x^2 + 2^x + 6 <= 134217727 will give x <= 11584.
uint32_t state_size = 6 +
state.optimizer.num_coef * state.optimizer.num_coef +
2 * state.optimizer.num_coef;
if(state_size > 134217727){
throw std::runtime_error(
"The product of number of independent variables and number of "
"categories cannot be larger than 11584.");
}
state.resize();
if (!args[3].isNull()) {
MultiResponseGLMState prev_state = args[3].getAs<ByteString>();
state = prev_state;
state.reset();
}
}
state << MutableMultiResponseGLMState::tuple_type(x, y);
return state.storage();
}
