ConfigSpace
ConfigSpace copied to clipboard
automl
Domain specific language for configuration spaces in Python/Cython. Useful for hyperparameter optimization and algorithm configuration.
```python from ConfigSpace.configuration_space import ConfigurationSpace, Configuration from ConfigSpace.hyperparameters import CategoricalHyperparameter from ConfigSpace.conditions import EqualsCondition, AndConjunction, OrConjunction from ConfigSpace.util import get_one_exchange_neighbourhood import numpy as np rng = np.random.RandomState(1) cs = ConfigurationSpace()...
TLDR: All hyperparameters have an internal float representation (yes, even categorical ones), which is either in the unit hypercube, or which is not in the unit hypercube, and either distributed...
Just a note that `get_neighbours` could be optimized in a few places. Currently we seem to generate a number at a time and then perform valid checks and then transform,...
`setup.py` only mentions the Linux platform, but we support Windows and OSX, too. Issue reported by @bastianzim in https://github.com/conda-forge/configspace-feedstock/pull/17
`Configuration.get()` always returns None if key is not in CS regardless of the provided default value. I have to use `config.get_dictionary.get()` to perform the expected behavior. ConfigSpace version: 0.4.20, 0.4.21,...
Is there a way to have a hierarchicaly structured configuration space such include arrays or dicts of scalar variables? For example to group say 5 float hyperparameters in a single...
Currently, it is not documented that `get_neighbors` operates in the unit hypercube. Also, as it operates in the unit hypercube, the `value` argument will always be of type `float`.
Similar to #232 we do not consider the default value for the hash, which should be fixed.
The typing for `is_legal` in `HyperParameter` is quite open and in general each subclass of `HyperParameter` is quite "open" with how it treats them, some checking for types explicitly while...
This would allow for * faster compilation of individual hyperparameters * easier refactor of the unit tests to use pytest style * easier overview of the code (the current files...