public function test_scenario1()
{
$data = array(array("predictions" => "./data/predictions_c.json", "method" => 0, "prediction" => "a", "confidence" => 0.450471270879), array("predictions" => "./data/predictions_c.json", "method" => 1, "prediction" => "a", "confidence" => 0.552021302649), array("predictions" => "./data/predictions_c.json", "method" => 2, "prediction" => "a", "confidence" => 0.403632421178), array("predictions" => "./data/predictions_r.json", "method" => 0, "prediction" => 1.55555556667, "confidence" => 0.400079152063), array("predictions" => "./data/predictions_r.json", "method" => 1, "prediction" => 1.59376845074, "confidence" => 0.248366474212), array("predictions" => "./data/predictions_r.json", "method" => 2, "prediction" => 1.55555556667, "confidence" => 0.400079152063));
foreach ($data as $item) {
print "\nSuccessfully computing predictions combinations\n";
$predictions = json_decode(file_get_contents($item["predictions"]));
print "Given I create a MultiVote for the set of predictions in file " . $item["predictions"] . "\n";
$multivote = new MultiVote($predictions);
print "When I compute the prediction with confidence using method " . $item["method"] . "\n";
$combined_results = $multivote->combine($item["method"], true);
print "And I compute the prediction without confidence using method " . $item["method"] . "\n";
$combined_results_no_confidence = $multivote->combine($item["method"]);
if ($multivote->is_regression()) {
print "Then the combined prediction is " . $item["prediction"] . "\n";
$this->assertEquals(round($combined_results[0], 6), round($item["prediction"], 6));
print "And the combined prediction without confidence is " . $item["prediction"] . "\n";
$this->assertEquals(round($combined_results_no_confidence, 6), round($item["prediction"], 6));
} else {
print "Then the combined prediction is " . $item["prediction"] . "\n";
$this->assertEquals($combined_results[0], $item["prediction"]);
print "And the combined prediction without confidence is " . $item["prediction"] . "\n";
$this->assertEquals($combined_results_no_confidence, $item["prediction"]);
}
print "And the confidence for the combined prediction is " . $item["confidence"] . "\n";
$this->assertEquals(round($combined_results[1], 6), round($item["confidence"], 6));
}
}
public function test_scenario1()
{
$data = array(array("predictions" => "./data/predictions_c.json", "method" => 0, "prediction" => "a", "confidence" => 0.450471270879), array("predictions" => "./data/predictions_c.json", "method" => 1, "prediction" => "a", "confidence" => 0.552021302649), array("predictions" => "./data/predictions_c.json", "method" => 2, "prediction" => "a", "confidence" => 0.403632421178), array("predictions" => "./data/predictions_r.json", "method" => 0, "prediction" => 1.55555556667, "confidence" => 0.400079152063), array("predictions" => "./data/predictions_r.json", "method" => 1, "prediction" => 1.59376845074, "confidence" => 0.248366474212), array("predictions" => "./data/predictions_r.json", "method" => 2, "prediction" => 1.55555556667, "confidence" => 0.400079152063));
foreach ($data as $item) {
$predictions = json_decode(file_get_contents($item["predictions"]));
$multivote = new MultiVote($predictions);
$combined_results = $multivote->combine($item["method"], true);
$combined_results_no_confidence = $multivote->combine($item["method"]);
if ($multivote->is_regression()) {
$this->assertEquals(round($combined_results[0], 6), round($item["prediction"], 6));
$this->assertEquals(round($combined_results_no_confidence, 6), round($item["prediction"], 6));
} else {
$this->assertEquals($combined_results[0], $item["prediction"]);
$this->assertEquals($combined_results_no_confidence, $item["prediction"]);
}
$this->assertEquals(round($combined_results[1], 6), round($item["confidence"], 6));
}
}
public function combine($method = MultiVote::DEFAULT_METHOD, $with_confidence = false, $add_confidence = false, $add_distribution = false, $add_count = false, $add_median = false, $add_min = false, $add_max = false, $options = null)
{
/*
Reduces a number of predictions voting for classification and
averaging predictions for regression.
method will determine the voting method (plurality, confidence
weighted, probability weighted or threshold).
If with_confidence is true, the combined confidence (as a weighted
average of the confidences of votes for the combined prediction)
will also be given.
*/
# there must be at least one prediction to be combined
if ($this->predictions == null) {
throw new Exception('No predictions to be combined.');
}
$method = array_key_exists(strval($method), $this->COMBINER_MAP) ? $this->COMBINER_MAP[strval($method)] : $this->COMBINER_MAP[MultiVote::DEFAULT_METHOD];
$keys = array_key_exists($method, $this->WEIGHT_KEYS) ? $this->WEIGHT_KEYS[$method] : null;
if ($keys != null) {
foreach ($keys as $key) {
foreach ($this->predictions as $prediction) {
if (!array_key_exists($key, $prediction)) {
throw new Exception('Not enough data to use the selected prediction method. Try creating your model anew.');
}
}
}
}
if ($this->is_regression()) {
$new_predictions = array();
foreach ($this->predictions as $prediction) {
if ($prediction->confidence == null) {
$prediction->confidence = 0;
}
$new_predictions[] = $prediction;
}
$this->predictions = $new_predictions;
if ($method == MultiVote::CONFIDENCE) {
return $this->error_weighted($with_confidence, $add_confidence, $add_distribution, $add_count, $add_median, $add_min, $add_max);
} else {
return $this->avg($this, $with_confidence, $add_confidence, $add_distribution, $add_count, $add_median, $add_min, $add_max);
}
} else {
$predictions = $this;
if ($method == MultiVote::THRESHOLD) {
if ($options == null) {
$options = array();
}
$predictions = $this->single_out_category($options);
} elseif ($method == MultiVote::PROBABILITY) {
$predictions = new MultiVote(array());
$predictions->predictions = $this->probability_weight();
}
return $predictions->combine_categorical(array_key_exists($method, $this->COMBINATION_WEIGHTS) ? $this->COMBINATION_WEIGHTS[$method] : null, $with_confidence, $add_confidence, $add_distribution, $add_count);
}
}
function predict($input_data, $by_name = true, $method = MultiVote::PLURALITY_CODE, $with_confidence = false, $add_confidence = false, $add_distribution = false, $add_count = false, $add_median = false, $add_min = false, $add_max = false, $options = null, $missing_strategy = Tree::LAST_PREDICTION, $median = false)
{
/*
Makes a prediction based on the prediction made by every model.
The method parameter is a numeric key to the following combination
methods in classifications/regressions:
0 - majority vote (plurality)/ average: PLURALITY_CODE
1 - confidence weighted majority vote / error weighted:
CONFIDENCE_CODE
2 - probability weighted majority vote / average:
PROBABILITY_CODE
3 - threshold filtered vote / doesn't apply:
THRESHOLD_CODE
*/
if (count($this->models_splits) > 1) {
$votes = new MultiVote(array());
$models = array();
$api = $this->api;
$order = 0;
foreach ($this->models_splits as $model_split) {
$models = array();
foreach ($model_split as $model_id) {
array_push($models, $api::retrieve_resource($model_id, $api::ONLY_MODEL));
}
$multi_model = new MultiModel($models, $this->api);
$votes_split = $multi_model->generate_votes($input_data, $by_name, $missing_strategy, $add_median || $median, $add_min, $add_max);
if ($median) {
foreach ($votes_split->predictions as $prediction) {
$prediction['prediction'] = $prediction['median'];
}
}
$votes->extend($votes_split->predictions);
}
#return $votes->combine($method, $with_confidence, $options);
} else {
# When only one group of models is found you use the
# corresponding multimodel to predict
$votes_split = $this->multi_model->generate_votes($input_data, $by_name, $missing_strategy, $add_median || $median, $add_min, $add_max);
$votes = new MultiVote($votes_split->predictions);
if ($median) {
$new_predictions = array();
foreach ($votes->predictions as $prediction) {
$prediction->prediction = $prediction->median;
array_push($new_predictions, $prediction);
}
$votes->predictions = $new_predictions;
}
}
return $votes->combine($method, $with_confidence, $add_confidence, $add_distribution, $add_count, $add_median, $add_min, $add_max, $options);
}
function _predict($input_data, $by_name = true, $method = MultiVote::PLURALITY_CODE, $with_confidence = false, $add_confidence = false, $add_distribution = false, $add_count = false, $add_median = false, $add_unused_fields = false, $add_min = false, $add_max = false, $options = null, $missing_strategy = Tree::LAST_PREDICTION, $median = false)
{
/*
Makes a prediction based on the prediction made by every model.
:param input_data: Test data to be used as input
:param by_name: Boolean that is set to true if field_names (as
alternative to field ids) are used in the
input_data dict
:param method: numeric key code for the following combination
methods in classifications/regressions:
0 - majority vote (plurality)/ average: PLURALITY_CODE
1 - confidence weighted majority vote / error weighted:
CONFIDENCE_CODE
2 - probability weighted majority vote / average:
PROBABILITY_CODE
3 - threshold filtered vote / doesn't apply:
THRESHOLD_CODE
The following parameter causes the result to be returned as a list
:param add_confidence: Adds confidence to the prediction
:param add_distribution: Adds the predicted node's distribution to the prediction
:param add_count: Adds the predicted nodes' instances to the prediction
:param add_median: Adds the median of the predicted nodes' distribution
to the prediction
:param add_min: Boolean, if true adds the minimum value in the
prediction's distribution (for regressions only)
:param add_max: Boolean, if true adds the maximum value in the
prediction's distribution (for regressions only)
:param add_unused_fields: Boolean, if true adds the information about
the fields in the input_data that are not
being used in the model as predictors.
:param options: Options to be used in threshold filtered votes.
:param missing_strategy: numeric key for the individual model's
prediction method. See the model predict
method.
:param median: Uses the median of each individual model's predicted
node as individual prediction for the specified
combination method.
*/
if (count($this->models_splits) > 1) {
$votes = new MultiVote(array());
$models = array();
$api = $this->api;
$order = 0;
foreach ($this->models_splits as $model_split) {
$models = array();
foreach ($model_split as $model_id) {
array_push($models, $api::retrieve_resource($model_id, $api::ONLY_MODEL));
}
$multi_model = new MultiModel($models, $this->api);
$votes_split = $multi_model->generate_votes($input_data, $by_name, $missing_strategy, $add_median || $median, $add_min, $add_max, $add_unused_fields);
if ($median) {
foreach ($votes_split->predictions as $prediction) {
$prediction['prediction'] = $prediction['median'];
}
}
$votes->extend($votes_split->predictions);
}
#return $votes->combine($method, $with_confidence, $options);
} else {
# When only one group of models is found you use the
# corresponding multimodel to predict
$votes_split = $this->multi_model->generate_votes($input_data, $by_name, $missing_strategy, $add_median || $median, $add_min, $add_max, $add_unused_fields);
$votes = new MultiVote($votes_split->predictions);
if ($median) {
$new_predictions = array();
foreach ($votes->predictions as $prediction) {
$prediction->prediction = $prediction->median;
array_push($new_predictions, $prediction);
}
$votes->predictions = $new_predictions;
}
}
$result = $votes->combine($method, $with_confidence, $add_confidence, $add_distribution, $add_count, $add_median, $add_min, $add_max, $options);
if ($add_unused_fields) {
$unused_fields = array_unique(array_keys($input_data));
foreach ($votes->predictions as $index => $prediction) {
$unused_fields = array_intersect($unused_fields, array_unique($prediction->unused_fields));
}
if (!is_array($result)) {
$result = array("prediction" => $result);
}
$result['unused_fields'] = $unused_fields;
}
return $result;
}
function generate_votes($input_data, $by_name = true, $missing_strategy = Tree::LAST_PREDICTION, $add_median = false, $add_min = false, $add_max = false)
{
/*
Generates a MultiVote object that contains the predictions
made by each of the models.
*/
$votes = new MultiVote(array());
$order = 0;
foreach ($this->models as $model) {
$prediction_info = $model->predict($input_data, $by_name, false, STDOUT, false, $missing_strategy, true, false, true, true, $add_median, false, $add_min, $add_max, null);
$votes->append($prediction_info);
}
return $votes;
}
