How to interpreter xgboost survival model result with and without output_margin

kpatel · May 29, 2019, 9:12pm

Hello Everyone,

I am new in xgboost. I am building survival model. When I get score with output_margin = False, all scores are between (0, 1). and when I get score with output_margin = True, I get all score below 0. I though since it is hazard ratio, range should be between (0, inf) if output_margin is False. And (-inf, inf) if we use output_margin = True.

If not case. how do I interpreter data?

Following image is density plot when output_margin is False

From,
Kush Patel

jmpanfil · November 12, 2019, 5:36pm

I am experiencing the same output. Can anyone comment on this?

hcho3 · November 12, 2019, 10:09pm

The link function is the exponential function:

github.com

dmlc/xgboost/blob/7663de956c37eb4dd528132214e68ba2851d9696/src/objective/regression_obj.cu#L336




    last_abs_y = abs_y;
    last_exp_p = exp_p;
  }
}
void PredTransform(HostDeviceVector<bst_float> *io_preds) override {
  std::vector<bst_float> &preds = io_preds->HostVector();
  const long ndata = static_cast<long>(preds.size()); // NOLINT(*)
#pragma omp parallel for schedule(static)
  for (long j = 0; j < ndata; ++j) {  // NOLINT(*)
    preds[j] = std::exp(preds[j]);
  }
}
void EvalTransform(HostDeviceVector<bst_float> *io_preds) override {
  PredTransform(io_preds);
}
bst_float ProbToMargin(bst_float base_score) const override {
  return std::log(base_score);
}
const char* DefaultEvalMetric() const override {
  return "cox-nloglik";

So the raw margin is an arbitrary real number in (-inf,inf), and the transformed output is in range (0, inf).