TransformerConstant is only supposed to handle np.number datatypes, but is used by Constant for union[int, float, str].

[jmtoepperwien]

The implementation of TransformerConstant.to_value uses np.full_like to return an array filled with the corresponding values. The dtype is given via type(self.value), but when Constant represents a str value, numpy uses U1 (unicode string of length 1) as the default datatype (at least for numpy version 1.26.3), truncating the value as a consequence.

The main problem here is, that the Transformer class is defined as only being able to handle np.number datatypes, but Constant uses it for all values anyways. A quick and dirty fix is to use 'U100' for np.full_like whenever the actual type is str, but this disregards the mismatch in the type hints.

As I've found this problem during debugging a given template, I don't have a minimal code example prepared, that exhibits this behavior.

[eddiebergman]

Repro:

from __future__ import annotations

from ConfigSpace import ConfigurationSpace

cs = ConfigurationSpace({"a": "a"})

s = cs.sample_configuration()
print(dict(s))
print(type(s["a"]))
print(s)

Issue fixed in #375