Feature Selection

See also

Concepts: Feature Selection

Prerequisites

This page uses the client and wait_for_task helpers defined in API Overview.

Feature selection identifies the most impactful features from a feature ideation using SHAP-based analysis. This is typically done automatically by the ideation pipeline, but you can also run it standalone to create custom selections with different parameters.

Run Feature Selection

response = client.post(
    "/feature_selection",
    json={
        "feature_ideation_id": feature_ideation_id,
        "target_feature_count": 100,
        "mode": "GENAI_BASED",
    },
)
task_id = response.json()["id"]
task = wait_for_task(client, task_id)

feature_selection_id = task.get("payload", {}).get("output_document_id")
print(f"Feature selection: {feature_selection_id}")

Parameters:

Parameter	Type	Required	Description
feature_ideation_id	string	Yes	ID of the feature ideation to select from
feature_selection_name	string	No	Custom name for the selection
mode	string	No	Selection mode: "GENAI_BASED" (default) or "RULE_BASED"
target_feature_count	integer	No	Target number of features to select (default: 50, max: 500)
use_relevance_score	boolean	No	Use semantic relevance scores in selection (default: true)
use_predictive_power_score	boolean	No	Use predictive power scores in selection (default: true)
remove_redundant_features	boolean	No	Remove highly correlated features (default: true)
remove_dictionary_and_vector	boolean	No	Exclude dictionary and vector features (default: true)
remove_low_added_value_features	boolean	No	Remove features with low marginal value (default: true)
keep_always_observation_features	boolean	No	Always keep features from the observation table (default: true)
rule	object	No	Rule-based selection parameters (only when mode is "RULE_BASED")
rule.top_n_overall	integer	No	Maximum features overall (default: 100)
rule.top_m_per_theme	integer	No	Maximum features per theme (default: 5)
rule.logic_operator	string	No	"OR" (default) or "AND" — how to combine top_n and top_m
observation_table_id	string	No	Observation table for SHAP evaluation
feature_ids	array	No	Restrict selection to specific feature IDs

Get Selection Results

response = client.get(f"/feature_selection/{feature_selection_id}")
selection = response.json()

print(f"Candidates: {selection['nb_candidates']}")
print(f"Selected: {selection['nb_selected']}")

Response fields:

Field	Type	Description
id	string	Feature selection ID
nb_candidates	integer	Number of candidate features evaluated
nb_selected	integer	Number of features selected
feature_ids	array	IDs of selected features
data	array	Per-feature results with feature_name, selection_rank, selection_rationale
signal_range	string	Signal range description

Create Feature List from Selection

Create a feature list containing the selected features:

response = client.post(f"/feature_selection/{feature_selection_id}/feature_list")
task_id = response.json()["id"]
task = wait_for_task(client, task_id)

feature_list_id = task.get("payload", {}).get("output_document_id")