Survival Analysis Output Configuration

Complete configuration guide for survival analysis and time-to-event modeling.

Overview

Survival outputs handle time-to-event analysis including:

  • Survival prediction - Time until event occurrence

  • Hazard modeling - Risk assessment over time

  • Censored data handling - Incomplete observation modeling

  • Medical prognosis - Patient outcome prediction

Quick Example

output_info:
  output_name: "patient_survival"
  output_type: "survival"
output_type_info:
  time_column: "survival_days"
  event_column: "death_observed"
model_config:
  model_type: "mlp-residual"
  model_init_config:
    fc_repr_dim: 64

Output Type Configuration

class eir.setup.schema_modules.output_schemas_survival.SurvivalOutputTypeConfig(
time_columns: list[str] = <factory>,
event_columns: list[str] = <factory>,
num_durations: int = 10,
loss_function: Literal['NegativeLogLikelihood',
'CoxPHLoss']='NegativeLogLikelihood',
max_duration: None | float = None,
label_parsing_chunk_size: None | int = None,
)

Basic configuration for survival analysis output.

Parameters:

  • time_columns – The names of the columns in the label file that contain the time-to-event or censoring time. Each time column should be aligned with the corresponding event column at the same index in event_columns.

  • event_columns – The names of the columns in the label file that indicate whether an event occurred (1) or the observation was censored (0). Each event column should be aligned with the corresponding time column at the same index in time_columns.

  • num_durations – The number of discrete time intervals to use in the model. This determines the size of the output layer per event.

  • loss_function – The loss function to use for training the survival model.

  • max_duration – The maximum duration to consider. Times beyond this will be censored at this point. If None, use the maximum observed time.

  • label_parsing_chunk_size – Number of rows to process at a time when loading the input_source. Useful when RAM is limited.

Output Module Configuration

Survival analysis typically uses tabular-based output modules. See Tabular Output Configuration for detailed configuration options of the underlying architectures.