def interpret_softmax_expr(self, expr, **kwargs):
if self.softmax_function_name is NotImplemented:
return self._do_interpret(fallback_expressions.softmax(expr.exprs),
**kwargs)
self.with_vectors = True
self.with_math_module = True
nested = [self._do_interpret(expr, **kwargs) for expr in expr.exprs]
return self._cg.function_invocation(self.softmax_function_name,
self._cg.vector_init(nested))
def _assemble_multi_class_output(self, estimator_params):
# Multi-class output is calculated based on discussion in
# https://github.com/dmlc/xgboost/issues/1746#issuecomment-295962863
# and the enhancement to support boosted forests in XGBoost.
splits = _split_estimator_params_by_classes(
estimator_params, self._output_size,
self.multiclass_params_seq_len)
base_score = self._base_score
exprs = [
self._assemble_single_output(e, base_score=base_score, split_idx=i)
for i, e in enumerate(splits)
]
proba_exprs = fallback_expressions.softmax(exprs)
return ast.VectorVal(proba_exprs)
def _multi_class_convert_output(self, exprs):
return fallback_expressions.softmax(exprs)