logml.survival_analysis.extractors.cox

Classes

CoxSAExtractor(cfg, sa_setup, global_params)

Runs Cox (backward + forward paths) analysis, saves the result model.

class logml.survival_analysis.extractors.cox.CoxSAParams

Bases: pydantic.main.BaseModel

Defines hyperparams for Cox BF method.

Show JSON schema

{
   "title": "CoxSAParams",
   "description": "Defines hyperparams for Cox BF method.",
   "type": "object",
   "properties": {
      "cox_model_penalizer": {
         "title": "Cox Model Penalizer",
         "description": "Defines a penalizer for CoxPHFitter model (lifelines package). Helps to avoid issues while fitting due to multicollinearity.",
         "default": 0.01,
         "type": "number"
      },
      "normalize_numericals": {
         "title": "Normalize Numericals",
         "description": "Whether numerical values should be normalized before fitting Cox model. Otherwise numerical values are to be binarized using thresholds found either via KaplanMeier or OptimalCutOff methods.",
         "default": false,
         "type": "boolean"
      },
      "univariate_p_value_threshold": {
         "title": "Univariate P Value Threshold",
         "description": "Defines a threshold that is used to select features for Cox Backward-Forward method from the summary table produced by OptimalCutOff method (applied to FDR corrected p-values).",
         "default": 0.05,
         "type": "number"
      },
      "covariate_p_value_threshold": {
         "title": "Covariate P Value Threshold",
         "description": "Defines a threshold that is used to check feature significance based on Cox model summary.",
         "default": 0.05,
         "type": "number"
      },
      "backward_path_single_removal_threshold": {
         "title": "Backward Path Single Removal Threshold",
         "default": 150,
         "type": "integer"
      },
      "backward_path_batch_size": {
         "title": "Backward Path Batch Size",
         "default": 30,
         "type": "integer"
      },
      "backward_path_batch_p_value": {
         "title": "Backward Path Batch P Value",
         "default": 0.7,
         "type": "number"
      }
   }
}

Fields

• backward_path_batch_p_value (float)

• backward_path_batch_size (int)

• backward_path_single_removal_threshold (int)

• covariate_p_value_threshold (float)

• cox_model_penalizer (float)

• normalize_numericals (bool)

• univariate_p_value_threshold (float)

field cox_model_penalizer: float = 0.01

Defines a penalizer for CoxPHFitter model (lifelines package). Helps to avoid issues while fitting due to multicollinearity.

field normalize_numericals: bool = False

Whether numerical values should be normalized before fitting Cox model. Otherwise numerical values are to be binarized using thresholds found either via KaplanMeier or OptimalCutOff methods.

field univariate_p_value_threshold: float = 0.05

Defines a threshold that is used to select features for Cox Backward-Forward method from the summary table produced by OptimalCutOff method (applied to FDR corrected p-values).

field covariate_p_value_threshold: float = 0.05

Defines a threshold that is used to check feature significance based on Cox model summary.

field backward_path_single_removal_threshold: int = 150

field backward_path_batch_size: int = 30

field backward_path_batch_p_value: float = 0.7

class logml.survival_analysis.extractors.cox.CoxSAExtractor(cfg: GlobalConfig, sa_setup: SurvivalAnalysisSetup, global_params: dict, logger=None)

Bases: logml.survival_analysis.extractors.base.BaseSAExtractor

Runs Cox (backward + forward paths) analysis, saves the result model.

LABEL = 'cox'

CONFIG_CLASS

alias of logml.survival_analysis.extractors.cox.CoxSAParams

run()

Three-step workflow:

• select initial features for Cox - all significant features

(based on univariate analysis) - backward path - sequentially removing features until we get significant model (all features are significant) - forward path - trying to add removed features back