/**
* GI that generates data in clusters, using a specified distribution for each
* cluster.
*
* This GI requires the following template arguments:
* - 'n' or 0
* The number of tuples to generate. Note: this value is per task.
* The total number of tuples generated will be n_tasks * n
* - 'centers' or 1
* A list of configuration for the centers.
*
* The following template arguments are optional:
* - 'outputs'
* If the outputs of the GI are not given implicitly, they can be
* specified in this template argument. The number of dimensions will
* be determined by the number of outputs.
*
* All output types must be numeric real types. The default type for
* outputs is DOUBLE.
* - 'dist.lib' = 'std'
* Which library to use for generating distributions.
* Valid options are:
* - std
* - boost
* - 'seed' = null
* The seed to be used for the random number generator. This seed will
* be used to generate the seed for each task, and different runs with
* the same seed will produce the same data.
* - 'compute.sets' = 1
* The number of sets of tuples to compute at once.
*
* Each center configuration is a functor with the form:
* dist_name(args)
*
* The following distributions are supported:
* { Uniform Distributions }
* - uniform(a = 0, b = 1)
*
* { Normal Distributions }
* - normal(mean = 0.0, std_dev = 1.0) [ synonyms: gaussian ]
* - inverse_gaussian(mean = 1, shape = 1) [ synonyms: inverse_normal ]
*
* { Bernoulli Distributions }
* - binomial(t = 1, p = 0.5)
* - negative_binomial(k = 1, p = 0.5)
*
* { Poisson Distributions }
* - exponential( lambda = 1 )
* - gamma(alpha = 1, beta = 1) [ synonyms: Gamma ]
*/
function ClusterGen(array $t_args, array $outputs)
{
$sys_headers = ['array', 'cinttypes'];
$user_headers = [];
$libraries = [];
if (\count($outputs) == 0) {
grokit_assert(array_key_exists('outputs', $t_args), 'ClusterGen: No outputs specified');
$count = 0;
foreach ($t_args['outputs'] as $type) {
if (is_identifier($type)) {
$type = lookupType($type);
}
grokit_assert(is_datatype($type), 'ClusterGen: Non data-type ' . $type . ' given as output');
$name = 'output' . $count++;
$outputs[$name] = $type;
}
}
foreach ($outputs as $name => &$type) {
if (is_null($type)) {
$type = lookupType('base::DOUBLE');
} else {
grokit_assert($type->is('real'), 'ClusterGen: Non-real datatype ' . $type . ' given as output');
}
}
$myOutputs = [];
foreach ($outputs as $name => $type) {
$myOutputs[$name] = $type;
}
$tSize = \count($outputs);
$seed = get_default($t_args, 'seed', null);
if ($seed !== null) {
grokit_assert(is_int($seed), 'ClusterGen: Seed must be an integer or null.');
} else {
$user_headers[] = 'HashFunctions.h';
}
$distLib = get_default($t_args, 'dist.lib', 'std');
$distNS = '';
switch ($distLib) {
case 'std':
$sys_headers[] = 'random';
$distNS = 'std';
break;
case 'boost':
$sys_headers[] = 'boost/random.hpp';
$distNS = 'boost::random';
$libraries[] = 'boost_random-mt';
if ($seed === null) {
// Need random_device
$sys_headers[] = 'boost/random/random_device.hpp';
$libraries[] = 'boost_system-mt';
}
break;
default:
grokit_error('ClusterGen: Unknown RNG library ' . $distLib);
}
$distRNG = 'mt19937';
$RNGtype = $distNS . '::' . $distRNG;
$nTuples = get_first_key($t_args, ['n', '0']);
grokit_assert(is_int($nTuples), 'ClusterGen: the number of tuples to be produced must be an integer.');
$centers = get_first_key($t_args, ['centers', 1]);
grokit_assert(is_array($centers), 'ClusterGen: centers must be an array of functors');
$handleDist = function ($name, $args, $oType) use($distNS) {
$distName = '';
$distArgs = [];
switch ($name) {
case 'gaussian':
case 'normal':
$distName = $distNS . '::' . 'normal_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 2, 'ClusterGen: Normal distribution takes at most 2 arguments, ' . \count($args) . ' given');
$mean = get_default($args, ['mean', 0], 0.0);
$sigma = get_default($args, ['std_dev', 'sigma', 1], 1.0);
grokit_assert(is_numeric($mean), 'ClusterGen: mean parameter of binomial distribution must be a real number.');
grokit_assert(is_numeric($sigma), 'ClusterGen: sigma parameter of binomial distribution must be a real number.');
$mean = floatval($mean);
$sigma = floatval($sigma);
$distArgs = [$mean, $sigma];
break;
case 'binomial':
$distName = $distNS . '::' . 'binomial_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 2, 'ClusterGen: Binomial distribution takes at most 2 arguments, ' . \count($args) . ' given');
$t = get_default($args, ['t', 0], 1);
$p = get_default($args, ['p', 1], 0.5);
grokit_assert(is_int($t), 'ClusterGen: t parameter of binomial distribution must be an integer.');
grokit_assert(is_numeric($p), 'ClusterGen: p parameter of binomial distribution must be a real number.');
$p = floatval($p);
grokit_assert($p >= 0 && $p <= 1, 'ClusterGen: p parameter of binomial distribution must be in the range [0, 1]');
grokit_assert($t >= 0, 'ClusterGen: t parameter of binomial distribution must be in the range [0, +inf)');
$distArgs = [$t, $p];
break;
case 'negative_binomial':
$distName = $distNS . '::' . 'negative_binomial_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 2, 'ClusterGen: Negative Binomial distribution takes at most 2 arguments, ' . \count($args) . ' given');
$k = get_default($args, ['k', 0], 1);
$p = get_default($args, ['p', 1], 0.5);
grokit_assert(is_int($k), 'ClusterGen: k parameter of binomial distribution must be an integer.');
grokit_assert(is_numeric($p), 'ClusterGen: p parameter of binomial distribution must be a real number.');
$p = floatval($p);
grokit_assert($p > 0 && $p <= 1, 'ClusterGen: p parameter of negative binomial distribution must be in the range (0, 1]');
grokit_assert($k > 0, 'ClusterGen: k parameter of negative binomial distribution must be in the range (0, +inf)');
$distArgs = [$k, $p];
break;
case 'inverse_gaussian':
case 'inverse_normal':
grokit_assert(\count($args) <= 2, 'ClusterGen: Inverse Gaussian distribution takes at most 2 arguments, ' . \count($args) . ' given');
$mean = get_default($args, ['mean', 0], 1);
$shape = get_default($args, ['shape', 1], 1);
grokit_assert(is_numeric($mean), 'ClusterGen: mean parameter of inverse gaussian distribution must be a real number.');
grokit_assert(is_numeric($shape), 'ClusterGen: shape parameter of inverse gaussian distribution must be a real number.');
$mean = floatval($mean);
$shape = floatval($shape);
grokit_assert($mean > 0, 'ClusterGen: mean of inverse gaussian distribution must be in range (0, inf)');
grokit_assert($shape > 0, 'ClusterGen: shape of inverse gaussian distribution must be in range (0, inf)');
$gen_args = ['output' => $oType, 'ns' => $distNS];
$distName = strval(lookupResource('datagen::InverseGaussianGen', $gen_args));
$distArgs = [$mean, $shape];
break;
case 'uniform':
$distName = $distNS . '::' . 'uniform_real_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 2, 'ClusterGen: Uniform distribution takes at most 2 arguments, ' . \count($args) . ' given');
$a = get_default($args, ['a', 0], 0.0);
$b = get_default($args, ['b', 1], 1.0);
grokit_assert(is_numeric($a), 'ClusterGen: `a` parameter of uniform distribution must be a real number.');
grokit_assert(is_numeric($b), 'ClusterGen: `b` parameter of uniform distribution must be a real number.');
$a = floatval($a);
$b = floatval($b);
grokit_assert($b >= $a, 'ClusterGen: `b` parameter of uniform distribution must be >= the `a` parameter.');
$distArgs = [$a, $b];
break;
case 'exponential':
$distName = $distNS . '::' . 'exponential_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 1, 'ClusterGen: Exponential distribution takes at most 1 argument.');
$lambda = get_default($args, ['lambda', 0], 1.0);
grokit_assert(is_numeric($lambda), 'ClusterGen: `lambda` parameter of exponential distribution must be a real number.');
$lambda = floatval($lambda);
grokit_assert($lambda > 0, 'ClusterGen: `lambda` parameter of exponential distribution must be in range (0, +inf).');
$distArgs = [$lambda];
break;
case 'gamma':
case 'Gamma':
$distName = $distNS . '::' . 'gamma_distribution<' . $oType . '>';
grokit_assert(\count($args) <= 2, 'ClusterGen: Gamma distribution takes at most 2 arguments.');
$alpha = get_default($args, ['alpha', 0], 1.0);
$beta = det_default($args, ['beta', 1], 1.0);
grokit_assert(is_numeric($alpha), 'ClusterGen: `alpha` parameter of gamma distribution must be a real number.');
grokit_assert(is_numeric($beta), 'ClusterGen: `beta` parameter of gamma distribution must be a real number.');
$alpha = floatval($alpha);
$beta = floatval($beta);
$distArgs = [$alpha, $beta];
break;
default:
grokit_error('ClusterGen: Unknown distribution ' . $name . ' given for center');
}
return [$distName, $distArgs];
};
$dists = [];
$distArgs = [];
$count = 0;
$oType = '';
$nCenters = 1;
reset($outputs);
foreach ($centers as $val) {
$cluster = $val;
if (is_functor($val)) {
$cluster = [$val];
} else {
if (is_array($val)) {
$nCenters = lcm($nCenters, \count($val));
} else {
grokit_error('ClusterGen: center descriptions must be functors or list of functors');
}
}
$curDist = [];
$curDistArgs = [];
$curDistName = 'distribution' . $count++;
$oType = strval(current($outputs));
$iCount = 0;
foreach ($cluster as $functor) {
grokit_assert(is_functor($functor), 'ClusterGen: center description must be a functor');
$vName = $curDistName . '_' . $iCount++;
$ret = $handleDist($functor->name(), $functor->args(), $oType);
$curDist[$vName] = $ret[0];
$curDistArgs[$vName] = $ret[1];
}
next($outputs);
$dists[$curDistName] = $curDist;
$distArgs[$curDistName] = $curDistArgs;
}
// Determine the default number of sets to compute at a time.
// We want to generate either $nTuples or 10,000 tuples, depending on which
// is less.
$defaultSetsTarget = min($nTuples, 10000);
$setsToTarget = intval(ceil($defaultSetsTarget / $nCenters));
$computeSets = get_default($t_args, 'compute.sets', $setsToTarget);
grokit_assert(is_int($computeSets) && $computeSets > 0, 'ClusterGen: compute.sets must be a positive integer, ' . $computeSets . ' given');
$className = generate_name('ClusterGen');
// For some BIZZARE reason, the $outputs array was getting modified while
// traversing over the $dists array. Making a deep copy of the outputs and
// then reassigning it seems to fix the issue.
$outputs = $myOutputs;
?>
class <?php
echo $className;
?>
{
// The number of tuples to produce per task
static constexpr size_t N = <?php
echo $nTuples;
?>
;
static constexpr size_t CacheSize = <?php
echo $computeSets * $nCenters;
?>
;
// Typedefs
typedef std::tuple<<?php
echo array_template('{val}', ', ', $outputs);
?>
> Tuple;
typedef std::array<Tuple, CacheSize> TupleArray;
typedef TupleArray::const_iterator TupleIterator;
typedef <?php
echo $RNGtype;
?>
RandGen;
// Number of tuples produced.
uintmax_t count;
// Cache a number of outputs for efficiency
TupleArray cache;
TupleIterator cacheIt;
// Random number generator
RandGen rng;
// Distributions
<?php
// This is the section causing issues.
foreach ($dists as $name => $list) {
foreach ($list as $vName => $type) {
?>
<?php
echo $type;
?>
<?php
echo $vName;
?>
;
<?php
}
// foreach distribution
}
// foreach cluster set
?>
// Helper function to generate tuples.
void GenerateTuples(void) {
<?php
$tIndex = 0;
foreach ($dists as $name => $list) {
$lCenters = \count($list);
// $nCenters has been defined to be the LCM of the number of centers in
// any column, so $lCenter is guaranteed to divide evenly into
// CacheSize
?>
for( size_t index = 0; CacheSize > index; index += <?php
echo $lCenters;
?>
) {
<?php
$index = 0;
foreach ($list as $vName => $type) {
?>
std::get<<?php
echo $tIndex;
?>
>(cache[index + <?php
echo $index;
?>
]) = <?php
echo $vName;
?>
(rng);
<?php
$index++;
}
// foreach value in tuple
?>
}
<?php
$tIndex++;
}
// foreach distribution
?>
cacheIt = cache.cbegin();
}
public:
// Constructor
<?php
echo $className;
?>
( GIStreamProxy & _stream ) :
cache()
, cacheIt()
, count(0)
, rng()
<?php
foreach ($dists as $name => $list) {
foreach ($list as $vName => $type) {
?>
, <?php
echo $vName;
?>
(<?php
echo implode(', ', $distArgs[$name][$vName]);
?>
)
<?php
}
// foreach distribution
}
// foreach cluster set
?>
{
<?php
if (is_null($seed)) {
?>
<?php
echo $distNS;
?>
::random_device rd;
<?php
}
// if seed is null
?>
RandGen::result_type seed = <?php
echo is_null($seed) ? 'rd()' : "CongruentHash({$seed}, _stream.get_id() )";
?>
;
rng.seed(seed);
cacheIt = cache.cend();
}
// Destructor
~<?php
echo $className;
?>
(void) { }
bool ProduceTuple(<?php
echo typed_ref_args($outputs);
?>
) {
if( N > count ) {
if( cacheIt == cache.cend() ) {
GenerateTuples();
}
<?php
$tIndex = 0;
foreach ($outputs as $name => $type) {
?>
<?php
echo $name;
?>
= std::get<<?php
echo $tIndex;
?>
>(*cacheIt);
<?php
$tIndex++;
}
// foreach output
?>
++cacheIt;
++count;
return true;
}
else {
return false;
}
}
};
<?php
return array('kind' => 'GI', 'name' => $className, 'output' => $outputs, 'system_headers' => $sys_headers, 'user_headers' => $user_headers, 'libraries' => $libraries);
}
function hashComplex($val)
{
$hasher = hash_init('sha256');
if (is_array($val)) {
hash_update($hasher, '[');
// Ensure array is sorted by keys
ksort($val);
foreach ($val as $name => $v) {
hash_update($hasher, $name);
hash_update($hasher, '=>');
hash_update($hasher, hashComplex($v));
}
hash_update($hasher, ']');
} else {
if (is_gla($val)) {
hash_update($hasher, 'gla');
hash_update($hasher, $val->hash());
} else {
if (is_gf($val)) {
hash_update($hasher, 'gf');
hash_update($hasher, $val->hash());
} else {
if (is_gt($val)) {
hash_update($hasher, 'gt');
hash_update($hasher, $val->hash());
} else {
if (is_gist($val)) {
hash_update($hasher, 'gist');
hash_update($hasher, $val->hash());
} else {
if (is_gi($val)) {
hash_update($hasher, 'gi');
hash_update($hasher, $val->hash());
} else {
if (is_datatype($val)) {
hash_update($hasher, 'datatype');
hash_update($hasher, $val->hash());
} else {
if (is_functor($val)) {
hash_update($hasher, 'functor');
hash_update($hasher, $val->hash());
} else {
if (is_identifier($val)) {
hash_update($hasher, 'identifier');
hash_update($hasher, $val->hash());
} else {
if (is_attribute($val)) {
hash_update($hasher, 'attribute');
hash_update($hasher, strval($val));
} else {
if (is_int($val) || is_float($val) || is_string($val) || is_bool($val)) {
hash_update($hasher, gettype($val));
hash_update($hasher, $val);
} else {
if (is_null($val)) {
hash_update($hasher, 'null');
} else {
$valType = is_object($val) ? get_class($val) : gettype($val);
grokit_logic_error('Unable to hash unknown type ' . $valType);
}
}
}
}
}
}
}
}
}
}
}
}
return hash_final($hasher);
}
