selector.selection_features

This module contains feature generation functions.

Classes

FeatureGenerator([logger])

Generate features necessary to evaluate configurations.

class selector.selection_features.FeatureGenerator(logger=None)

Bases: object

Generate features necessary to evaluate configurations.

Parameters:

logger (logging.Logger) – Logging object.

avg_dist_evals(suggests, evals, psetting)

Average distance to all points so far evaluated.

Parameters:

• suggestions (list) – Suggested points.

• evals (list) – Already evaluated points.

• psetting (object) – Scenario parameters.

Returns:

div_feats – Average distances to all already evaluated points.

Return type:

list

avg_dist_rel(suggests, evals, psetting, generators)

Average distances to relatives.

Parameters:

• suggests (list) – Suggested points.

• evals (list) – Already evaluated points.

• psetting (object) – Scenario parameters.

• generators (list) – Available generators.

Returns:

div_feats – Computed features of suggested points.

Return type:

list

avg_dist_sel(suggests, psetting)

Average distance to points in the current selection.

Parameters:

• suggestions (list) – Suggested points.

• psetting (object) – Scenario parameters.

Returns:

div_feats – Average distances to points in the current selection.

Return type:

list

avg_rel_evals_qual(suggestions, data, nce, results, cot, generators)

Average quality of relatives so far evaluated.

Parameters:

• suggestions (list) – Suggested points.

• data (object) – Contains historic performance data.

• nce (int) – Number of all configs evaluated.

• results (dict) – Qualities of configurations.

• cot (float) – Cut off time for tournaments (i.e. time limit).

• generators (list) – All possible generators.

Returns:

div_feats – Computed features of suggested points.

Return type:

list

best_rel_evals_qual(suggestions, data, generators, results, cot)

Best target value relatives so far evaluated.

Parameters:

• suggestions (list) – Suggested points.

• data (object) – Contains historic performance data.

• generators (list) – All possible generators.

• results (dict) – Qualities of configurations.

• cot (float) – Cut off time for tournaments (i.e. time limit).

Returns:

div_feats – Computed features of suggested points.

Return type:

list

diff_pred_real_qual(suggestions, data, predicted_quals, results)

Difference of predicted & real qual. of relatives evaluated so far.

Parameters:

• suggestions (list) – Suggested points.

• data (object) – Contains historic performance data.

• predicted_quals (list of lists) – Predicted performance/quality for suggested configurations.

• results (dict) – Qualities of configurations.

Returns:

div_feats – Computed features of suggested points.

Return type:

list

diversity_feature_gen(suggestions, data, results, cot, psetting, predicted_quals, evaluated)

Generate diversity features.

Parameters:

• suggestions (list) – Suggested configurations.

• data (object) – Contains historic data.

• results (dict) – Qualities of configurations.

• cot (float) – Cut off time for tournaments.

• psetting (object) – Scenario parameters.

• predicted_quals (list) – Predicted qualities of points evaluated so far.

• evaluated (list) – All evaluated points so far.

• sm (object) – Initialized Surrogates.SurrogateManager().

Returns:

div_feats – Diversity features.

Return type:

list

dynamic_feature_gen(suggestions, data, predicted_quals, sm, cot, results, next_instance_set)

Generate dynamic features.

Parameters:

• suggestions (list) – Suggested configurations.

• data (object) – Contains historic data.

• predicted_quals (list of lists) – Predicted performance/quality for suggested configurations.

• sm (object) – Surrogates.SurrogateManager().

• cot (int) – Cut off time (i.e. time limit).

• results (list) – Results of points evaluated so far.

Returns:

dyn_feats – Dynamic features.

Return type:

list

expected_improve(suggs, sm, cot, surr, next_instance_set)

Probability of quality of points to improve.

Parameters:

• suggests (list) – Suggested points.

• sm (object) – Surrogates.SurrogateManager().

• cot (int) – Cut off time (i.e. time limit).

• surr (str) – Which surrogate to use.

Returns:

dyn_feats – Computed features of suggested points.

Return type:

list

expected_qual(suggs, sm, cot, surr, next_instance_set)

Expected quality of points.

Parameters:

• suggests (list) – Suggested points.

• sm (object) – Surrogates.SurrogateManager().

• cot (int) – Cut off time (i.e. time limit).

• surr (str) – Which surrogate to use.

• next_instance_set (list) – Next instances that will be run.

Returns:

dyn_feats – Computed features of suggested points.

Return type:

list

percent_rel_evals(suggestions, data, nce)

Percentage of relatives so far evaluated.

Parameters:

• suggestions (list) – Suggested points.

• data (object) – Contains historic performance data.

• nce (int) – Number of configuration evaluations.

Returns:

div_feats – Computed features of suggested points.

Return type:

list

prob_qual_improve(suggs, sm, cot, results, surr, next_instance_set)

Probability of quality of points to improve.

Parameters:

• suggests (list) – Suggested points.

• sm (object) – Surrogates.SurrogateManager().

• cot (int) – Cut off time (i.e. time limit).

• results (list) – Results of points evaluated so far.

• surr (str) – Which surrogate to use.

Returns:

dyn_feats – Computed features of suggested points.

Return type:

list

static_feature_gen(suggestions, epoch, max_epoch)

Generate static features.

Parameters:

• suggestions (list) – Suggested configurations.

• epoch (int) – Current epoch.

• max_epoch (int) – Total number of epochs.

Returns:

static_features – Static features.

Return type:

list

std_rel_evals_qual(suggestions, data, generators, results, cot)

Std of quality of relatives so far evaluated.

Parameters:

• suggestions (list) – Suggested points.

• data (object) – Contains historic performance data.

• generators (list) – All possible generators.

• results (dict) – Qualities of configurations.

• cot (float) – Cut off time for tournaments (i.e. time limit).

Returns:

div_feats – Computed features of suggested points.

Return type:

list

surr_votes(dyn_feats)

Multiply surr features to get agreement features.

Parameters:

dyn_feats (list of np.ndarray) – Dynamic features.

Returns:

dyn_feats – Extended dynamic features.

Return type:

list of np.ndarray

uncertainty_improve(suggs, sm, cot, surr, next_instance_set)

Probability of quality of points to improve.

Parameters:

• suggests (list) – Suggested points.

• sm (object) – Surrogates.SurrogateManager().

• cot (int) – Cut off time (i.e. time limit).

• surr (str) – Which surrogate to use.

Returns:

dyn_feats – Computed features of suggested points.

Return type:

list