Set additional parameters in algorithms (parameters which are not tuned)

[jmrichardson]

Hi, is it possible to send parameters to a specific model? I am currently just using lightgbm and would like to set the following: is_unbalance=True

Thank you

[pplonski]

@jmrichardson there is no such functionality, but I will add it.

[jmrichardson]

Great news! It would also be great to be able to set default parameters per model. I have a pretty unique dataset and the default parameters don't quite work well for initial start. Thank you

[pplonski]

@jmrichardson good idea. Looks that there should be option to set 3 types of parameters:

• additional parameters for algorithms (not tuned, just passed to the algorithm)
• default parameters for algorithm (that will be used in first model training),
• new range of parameters that will be optimized for algorithm

[jmrichardson]

Hi @pplonski , wanted to see if this is on the roadmap and when you think it might be available? Thanks for your consideration.

[pplonski]

Hey @jmrichardson, I will try to add this to the next release. But cant tell when it will be, I started to work on our new product: a notebook with no-code blocks which will make Python coding simple and available for many (the new product will have our AutoML).

[jmrichardson]

Hi, thank you for the fast reply! The new product I am sure will be very useful to the community! There is no rush, especially after testing the Optuna integration which handles most of my needs. Thanks again.

[alitirmizi23]

Any idea if we can do this now? with automl and having custom parameters per model?

[pplonski]

Hi @alitirmizi23! Currently, all my efforts are focused on the https://github.com/mljar/mercury framework for converting notebooks to web apps. I hope I can built sustainable bussiness around Mercury and use founds to develop MLJAR AutoML.

I do provide commercial support for MLJAR AutoML. I can implement this feature if needed by a commercial client. Licenses start at $1000/year.

The alternative is to implement this yourself or wait - from time to time I have interns that can code this.

[alitirmizi23]

Hey @pplonski, I understand. Maybe I'll try to implement myself. as far as license goes, we already have another enterprise automl framework. To your credit, i have found MLJAR Supervised to be one of the best open source autoML frameworks available

[pplonski]

@adrianblazeusz could you please take a look at this issue?