public function mlfindattributesAction()
{
$current_model = $model_info = $instance = $instances = $message = $tree_descriptor = $model_html = $config = '';
$possible_models = $possible_models_id = $other_models = array();
$jsonData = $jsonHeader = $jsonColumns = $jsonColor = '[]';
$jsonFAttrs = $jsonFAttrsHeader = '[]';
$mae = $rae = 0;
$must_wait = 'NO';
try {
$dbml = new \PDO($this->container->get('config')['db_conn_chain'], $this->container->get('config')['mysql_user'], $this->container->get('config')['mysql_pwd']);
$dbml->setAttribute(\PDO::ATTR_ERRMODE, \PDO::ERRMODE_EXCEPTION);
$dbml->setAttribute(\PDO::ATTR_EMULATE_PREPARES, false);
$db = $this->container->getDBUtils();
// FIXME - This must be counted BEFORE building filters, as filters inject rubbish in GET when there are no parameters...
$instructions = count($_GET) <= 1;
if (array_key_exists('dump', $_GET)) {
$dump = $_GET["dump"];
unset($_GET["dump"]);
}
if (array_key_exists('pass', $_GET)) {
$pass = $_GET["pass"];
unset($_GET["pass"]);
}
$this->buildFilters(array('current_model' => array('type' => 'selectOne', 'default' => null, 'label' => 'Model to use: ', 'generateChoices' => function () {
return array();
}, 'parseFunction' => function () {
$choice = isset($_GET['current_model']) ? $_GET['current_model'] : array("");
return array('whereClause' => '', 'currentChoice' => $choice);
}, 'filterGroup' => 'MLearning'), 'unseen' => array('type' => 'checkbox', 'default' => 1, 'label' => 'Predict with unseen atributes ⚠', 'parseFunction' => function () {
$choice = isset($_GET['unseen']) && !isset($_GET['unseen']) ? 0 : 1;
return array('whereClause' => '', 'currentChoice' => $choice);
}, 'filterGroup' => 'MLearning'), 'minexetime' => array('default' => 0), 'valid' => array('default' => 0), 'filter' => array('default' => 0), 'prepares' => array('default' => 1)));
$this->buildFilterGroups(array('MLearning' => array('label' => 'Machine Learning', 'tabOpenDefault' => true, 'filters' => array('current_model', 'unseen'))));
$param_names = array('bench', 'net', 'disk', 'maps', 'iosf', 'replication', 'iofilebuf', 'comp', 'blk_size', 'id_cluster', 'datanodes', 'vm_OS', 'vm_cores', 'vm_RAM', 'provider', 'vm_size', 'type', 'bench_type', 'hadoop_version', 'datasize', 'scale_factor');
// Order is important
$params = $this->filters->getFiltersSelectedChoices($param_names);
foreach ($param_names as $p) {
if (!is_null($params[$p]) && is_array($params[$p])) {
sort($params[$p]);
}
}
$learnParams = $this->filters->getFiltersSelectedChoices(array('current_model', 'unseen'));
$param_current_model = $learnParams['current_model'];
$unseen = $learnParams['unseen'] ? true : false;
$where_configs = $this->filters->getWhereClause();
$where_configs = str_replace("AND .", "AND ", $where_configs);
// compose instance
$model_info = MLUtils::generateModelInfo($this->filters, $param_names, $params, $unseen);
$instance = MLUtils::generateSimpleInstance($this->filters, $param_names, $params, $unseen);
$instances = MLUtils::generateInstances($this->filters, $param_names, $params, $unseen, $db);
// Model for filling
MLUtils::findMatchingModels($model_info, $possible_models, $possible_models_id, $dbml);
$current_model = '';
if (!is_null($possible_models_id) && in_array($param_current_model, $possible_models_id)) {
$current_model = $param_current_model;
}
// Other models for filling
$where_models = '';
if (!empty($possible_models_id)) {
$where_models = " WHERE id_learner NOT IN ('" . implode("','", $possible_models_id) . "')";
}
$result = $dbml->query("SELECT id_learner FROM aloja_ml.learners" . $where_models);
foreach ($result as $row) {
$other_models[] = $row['id_learner'];
}
if ($instructions) {
$result = $dbml->query("SELECT id_learner, model, algorithm FROM aloja_ml.learners");
foreach ($result as $row) {
$model_html = $model_html . "<li>" . $row['id_learner'] . " => " . $row['algorithm'] . " : " . $row['model'] . "</li>";
}
MLUtils::getIndexFAttrs($jsonFAttrs, $jsonFAttrsHeader, $dbml);
$this->filters->setCurrentChoices('current_model', array_merge($possible_models_id, array('---Other models---'), $other_models));
return $this->render('mltemplate/mlfindattributes.html.twig', array('fattrs' => $jsonFAttrs, 'header_fattrs' => $jsonFAttrsHeader, 'models' => $model_html, 'instructions' => 'YES'));
}
if (!empty($possible_models_id) || $current_model != "") {
$result = $dbml->query("SELECT id_learner, model, algorithm, CASE WHEN `id_learner` IN ('" . implode("','", $possible_models_id) . "') THEN 'COMPATIBLE' ELSE 'NOT MATCHED' END AS compatible FROM aloja_ml.learners");
foreach ($result as $row) {
$model_html = $model_html . "<li>" . $row['id_learner'] . " => " . $row['algorithm'] . " : " . $row['compatible'] . " : " . $row['model'] . "</li>";
}
if ($current_model == "") {
$query = "SELECT AVG(ABS(exe_time - pred_time)) AS MAE, AVG(ABS(exe_time - pred_time)/exe_time) AS RAE, p.id_learner FROM aloja_ml.predictions p, aloja_ml.learners l WHERE l.id_learner = p.id_learner AND p.id_learner IN ('" . implode("','", $possible_models_id) . "') AND predict_code > 0 ORDER BY MAE LIMIT 1";
$result = $dbml->query($query);
$row = $result->fetch();
$current_model = $row['id_learner'];
}
$config = $instance . '-' . $current_model . '-' . ($unseen ? 'U' : 'R');
$is_cached_mysql = $dbml->query("SELECT count(*) as total FROM aloja_ml.trees WHERE id_findattrs = '" . md5($config) . "'");
$tmp_result = $is_cached_mysql->fetch();
$is_cached = $tmp_result['total'] > 0;
$tmp_file = md5($config) . '.tmp';
$in_process = file_exists(getcwd() . '/cache/ml/' . md5($config) . '.lock');
$finished_process = $in_process && (int) shell_exec('ls ' . getcwd() . '/cache/ml/' . md5($config) . '-*.lock | wc -w ') == count($instances);
if (!$in_process && !$finished_process && !$is_cached) {
// Retrieve file model from DB
$query = "SELECT file FROM aloja_ml.model_storage WHERE id_hash='" . $current_model . "' AND type='learner';";
$result = $dbml->query($query);
$row = $result->fetch();
$content = $row['file'];
$filemodel = getcwd() . '/cache/ml/' . $current_model . '-object.rds';
$fp = fopen($filemodel, 'w');
fwrite($fp, $content);
fclose($fp);
// Run the predictor
exec('cd ' . getcwd() . '/cache/ml ; touch ' . md5($config) . '.lock ; rm -f ' . $tmp_file);
$count = 1;
foreach ($instances as $inst) {
exec(getcwd() . '/resources/queue -d -c "cd ' . getcwd() . '/cache/ml ; ../../resources/aloja_cli.r -m aloja_predict_instance -l ' . $current_model . ' -p inst_predict=\'' . $inst . '\' -v | grep -v \'Prediction\' >>' . $tmp_file . ' 2>/dev/null; touch ' . md5($config) . '-' . $count++ . '.lock" >/dev/null 2>&1 &');
}
}
$finished_process = (int) shell_exec('ls ' . getcwd() . '/cache/ml/' . md5($config) . '-*.lock | wc -w ') == count($instances);
if ($finished_process && !$is_cached) {
// Read results and dump to DB
$i = 0;
$token = 0;
$token_i = 0;
$query_pattern = "INSERT IGNORE INTO aloja_ml.predictions (\n\t\t\t\t\t\tid_exec,exe_time,bench,net,disk,maps,iosf,replication,iofilebuf,comp,blk_size,\n\t\t\t\t\t\tid_cluster,datanodes,vm_OS,vm_cores,vm_RAM,provider,vm_size,type,bench_type,hadoop_version,\n\t\t\t\t\t\tdatasize,scale_factor,\n\t\t\t\t\t\tnet_maxtxkbs,net_maxrxkbs,net_maxtxpcks,net_maxrxpcks,net_maxtxcmps,net_maxrxcmps,net_maxrxmscts,\n\t\t\t\t\t\tdisk_maxtps,disk_maxsvctm,disk_maxrds,disk_maxwrs,disk_maxrqsz,disk_maxqusz,disk_maxawait, disk_maxutil,\n\t\t\t\t\t\tpred_time,id_learner,instance,predict_code) VALUES ";
$query = $query_pattern;
if (($handle = fopen(getcwd() . '/cache/ml/' . $tmp_file, "r")) !== FALSE) {
while (($line = fgets($handle, 1000)) !== FALSE && $i < 1000) {
if ($line == '') {
break;
}
// Fetch Real Value
$inst_aux = preg_split("/\\s+/", $line);
$query_var = "SELECT AVG(exe_time) as AVG, id_exec, outlier FROM aloja_ml.predictions WHERE instance = '" . $inst_aux[1] . "' AND predict_code > 0";
$result = $dbml->query($query_var);
$row = $result->fetch();
$realexecval = is_null($row['AVG']) || $row['outlier'] == 2 ? 0 : $row['AVG'];
$realid_exec = is_null($row['id_exec']) || $row['outlier'] == 2 ? 0 : $row['id_exec'];
$query_var = "SELECT count(*) as num FROM aloja_ml.predictions WHERE instance = '" . $inst_aux[1] . "' AND id_learner = '" . $current_model . "'";
$result = $dbml->query($query_var);
$row = $result->fetch();
// Insert instance values
if ($row['num'] == 0) {
$token_i = 1;
$selected_instance = preg_replace('/,Cmp(\\d+),/', ',${1},', $inst_aux[1]);
$selected_instance = preg_replace('/,Cl(\\d+),/', ',${1},', $selected_instance);
if ($token > 0) {
$query = $query . ",";
}
$token = 1;
$query = $query . "('" . $realid_exec . "','" . $realexecval . "','" . str_replace(",", "','", $selected_instance) . "','" . $inst_aux[2] . "','" . $current_model . "','" . $inst_aux[1] . "','0') ";
}
$i++;
if ($i % 100 == 0 && $token_i > 0) {
if ($dbml->query($query) === FALSE) {
throw new \Exception('Error when saving into DB');
}
$query = $query_pattern;
$token = 0;
$token_i = 0;
}
}
if ($token_i > 0) {
if ($dbml->query($query) === FALSE) {
throw new \Exception('Error when saving into DB');
}
}
// Descriptive Tree
$tree_descriptor = shell_exec(getcwd() . '/resources/aloja_cli.r -m aloja_representative_tree -p method=ordered:dump_file="' . getcwd() . '/cache/ml/' . $tmp_file . '":output=nodejson -v 2> /dev/null');
$tree_descriptor = substr($tree_descriptor, 5, -2);
$tree_descriptor = str_replace("\\\"", "\"", $tree_descriptor);
$tree_descriptor = str_replace("desc:\"\"", "desc:\"---\"", $tree_descriptor);
$query = "INSERT INTO aloja_ml.trees (id_findattrs,id_learner,instance,model,tree_code) VALUES ('" . md5($config) . "','" . $current_model . "','" . $instance . "','" . $model_info . "','" . $tree_descriptor . "')";
if ($dbml->query($query) === FALSE) {
throw new \Exception('Error when saving tree into DB');
}
// remove remaining locks
shell_exec('rm -f ' . getcwd() . '/cache/ml/' . md5($config) . '*.lock');
// Remove temporal files
$output = shell_exec('rm -f ' . getcwd() . '/cache/ml/' . md5($config) . '.tmp');
$is_cached = true;
}
fclose($handle);
}
if (!$is_cached) {
$must_wait = 'YES';
if (isset($dump)) {
$dbml = null;
echo "1";
exit(0);
}
if (isset($pass)) {
$dbml = null;
return "1";
}
throw new \Exception('WAIT');
}
if (isset($pass) && $pass == 2) {
$dbml = null;
return "2";
}
// Fetch results and compose JSON
$header = array('Benchmark', 'Net', 'Disk', 'Maps', 'IO.SFS', 'Rep', 'IO.FBuf', 'Comp', 'Blk.Size', 'Cluster', 'Datanodes', 'VM.OS', 'VM.Cores', 'VM.RAM', 'Provider', 'VM.Size', 'Type', 'Bench.Type', 'Version', 'Datasize', 'Scale.Factor', 'Prediction', 'Observed');
$jsonHeader = '[{title:""}';
foreach ($header as $title) {
$jsonHeader = $jsonHeader . ',{title:"' . $title . '"}';
}
$jsonHeader = $jsonHeader . ']';
$query = "SELECT @i:=@i+1 as num, instance, AVG(pred_time) as pred_time, AVG(exe_time) as exe_time FROM aloja_ml.predictions AS e, (SELECT @i:=0) d WHERE id_learner='" . $current_model . "' " . $where_configs . " GROUP BY instance";
$result = $dbml->query($query);
$jsonData = '[';
foreach ($result as $row) {
if ($jsonData != '[') {
$jsonData = $jsonData . ',';
}
$jsonData = $jsonData . "['" . $row['num'] . "','" . implode("','", array_slice(explode(",", $row['instance']), 0, 21)) . "','" . $row['pred_time'] . "','" . $row['exe_time'] . "']";
}
$jsonData = $jsonData . ']';
foreach (range(1, 33) as $value) {
$jsonData = str_replace('Cmp' . $value, Utils::getCompressionName($value), $jsonData);
}
// Fetch MAE & RAE values
$query = "SELECT AVG(ABS(exe_time - pred_time)) AS MAE, AVG(ABS(exe_time - pred_time)/exe_time) AS RAE FROM aloja_ml.predictions AS e WHERE id_learner='" . md5($config) . "' AND predict_code > 0";
$result = $dbml->query($query);
$row = $result->fetch();
$mae = $row['MAE'];
$rae = $row['RAE'];
// Dump case
if (isset($dump)) {
echo "ID" . str_replace(array("[", "]", "{title:\"", "\"}"), array('', '', ''), $jsonHeader) . "\n";
echo str_replace(array('],[', '[[', ']]'), array("\n", '', ''), $jsonData);
$dbml = null;
exit(0);
}
if (isset($pass) && $pass == 1) {
$retval = "ID" . str_replace(array("[", "]", "{title:\"", "\"}"), array('', '', ''), $jsonHeader) . "\n";
$retval .= str_replace(array('],[', '[[', ']]'), array("\n", '', ''), $jsonData);
$dbml = null;
return $retval;
}
// Display Descriptive Tree
$query = "SELECT tree_code FROM aloja_ml.trees WHERE id_findattrs = '" . md5($config) . "'";
$result = $dbml->query($query);
$row = $result->fetch();
$tree_descriptor = $row['tree_code'];
} else {
if (isset($dump)) {
echo "-1";
exit(0);
}
if (isset($pass)) {
return "-1";
}
throw new \Exception("There are no prediction models trained for such parameters. Train at least one model in 'ML Prediction' section.");
}
} catch (\Exception $e) {
if ($e->getMessage() != "WAIT") {
$this->container->getTwig()->addGlobal('message', $e->getMessage() . "\n");
}
$jsonData = $jsonHeader = $jsonColumns = $jsonColor = '[]';
$instances = array();
if (isset($pass)) {
return "-2";
}
}
$dbml = null;
$return_params = array('instance' => $instance, 'jsonData' => $jsonData, 'jsonHeader' => $jsonHeader, 'fattrs' => $jsonFAttrs, 'header_fattrs' => $jsonFAttrsHeader, 'models' => $model_html, 'models_id' => $possible_models_id, 'other_models_id' => $other_models, 'current_model' => $current_model, 'mae' => $mae, 'rae' => $rae, 'must_wait' => $must_wait, 'instance' => $instance, 'instances' => implode("<br/> ", $instances), 'model_info' => $model_info, 'id_findattr' => md5($config), 'tree_descriptor' => $tree_descriptor);
$this->filters->setCurrentChoices('current_model', array_merge($possible_models_id, array('---Other models---'), $other_models));
return $this->render('mltemplate/mlfindattributes.html.twig', $return_params);
}
