-
Did you consider using more datasets?
And how about regression problems?
There is for example this benchmarking suite, accessible via the OpenML packages: [https://arxiv.org/abs/1708.03731](htt…
-
GLM generally requires the most feature engineering because it builds a linear model. This request is to perform the common feature engineering as a pre-processing step to train GLM when running it i…
-
We have a benchmark where we run lots of long jobs with default settings. It has now failed 3 times on a dataset called pc_krkopt. It looks like this
```AutoML progress: |████████████████████████████…
-
The `ConfigurationSpace` can be a source of randomness and unfortunatly only provides a `seed(...)` functionality to update it. I don't trust this for re-use across multiple benchmark runs. If there w…
-
AutoML is currently using default AUTO SE config for all its SE models.
We need to investigate if this is always the best choice, for example if a different metalearner algo could be picked in some s…
-
Similarly to `ECP-Candle` and `PINNBench` benchmarks some new benchmarks compatible with Multi-Objective Optimization (MOO) should be defined. They could be "extensions" of the already existing benchm…
-
Hello there,
We (@glouppe , @betatim and myself) have been working on https://github.com/scikit-optimize/scikit-optimize over the past few months and just made the first release. For the next release…
-
Currently `gym.Spaces` as well as the sample drawn from them are serialized to JSON in different ways producing duplicated code and incompatible formats at a different places of the program.
Known…
-
Consider turning off early stopping in the default DRF and XRT models in AutoML since we only have 50 trees. First, evaluate on some datasets and see how this affects performance. Let's benchmark to …
-
Really enjoy working with the package! Would it be possible to implement the use of custom metrics? Some AutoML frameworks like autosklearn support this while others like h2o do not. I think it would …