survmetr 0.1.0

Advanced predictive metrics for survival analysis

survmetr

Advanced predictive metrics for survival analysis, focused on Antolini's time-dependent concordance index.

Installation

pip install survmetr

For numba-accelerated computation:

pip install survmetr[numba]

Quick Start

import numpy as np
from sksurv.util import Surv
from survmetr import c_index_antolini

# Create survival data
y = Surv.from_arrays(
    event=[True, True, False, True, False],
    time=[1.0, 3.0, 4.0, 5.0, 6.0],
)
n_events = y['event'].sum()

# Prediction matrix: (n_samples, n_events) failure probabilities
estimate = np.random.rand(len(y), n_events)

# Compute Antolini's time-dependent C-index
c = c_index_antolini(estimate, y)
print(f"C-index: {c:.3f}")

# Get detailed statistics
result = c_index_antolini(estimate, y, return_all=True)
print(result)
# {'c-index': ..., 'concordant': ..., 'comparable': ...,
#  'tied_risk': ..., 'discordant': ..., 'numerator': ...}

Available Metrics

Function	Description
c_index_antolini	Antolini's time-dependent C-index (alias for vector impl)
c_index_antolini_vector	Vectorized NumPy implementation
c_index_antolini_sksurv	Implementation using scikit-survival internals
c_index_antolini_pycox	Numba-JIT parallel implementation
make_survival_scorer	Factory for time-dependent classification scorers
harrel_c_index_scorer	Harrell's standard C-index scorer
AntoliniCIndexVecScorer	Dataclass scorer for the vector implementation
split_y	Extract event/time from structured arrays

Tutorial

See Tutorial.ipynb for a detailed walkthrough.

License

This project is licensed under the GNU Lesser General Public License v3.0 - see the LICENSE file for details.