Dear All,
We have released a complete set of linear models. We also released some predefined models to go along with it. These models have a 1:1 correspondance with SciKit Learn models
(note we are still on version 1.5, we will be evaluating to 1.8 soon)
Lasso → Lasso — scikit-learn 1.8.0 documentation
Ridge → Ridge — scikit-learn 1.8.0 documentation
ElasticNet → ElasticNet — scikit-learn 1.8.0 documentation
LinearRegression → LinearRegression — scikit-learn 1.8.0 documentation
Investor Friendly
Please see the descriptions of the hyper parameters. We are trying to write them in an investor friendly way, specific to our domain (models for noisy financial data).
Coefficients
You can view the coefficients of linear models by training a Predictor and going to the Importance tab. This solution is a bit of a hack since training a predictor is usually the last step of an AI workflow. We're working on version 2.0 of AI Factor to clean up the workflow, and generally make AI Factor more usable.
Previous Version
The previous linear model was an unfortunate choice and caused a lot of confusion (see the posts below). It was based on ElasticNet, but switched to LinearRegression depending on the value of a hyperparameter. Lastly the old models used fit_intercept=true, which is the wrong choice for us since we use normalized data (we are not predicting actual future returns, just the z-scores). The old linear model will be deprecated at some point. For now it has just been renamed to "deprecated".
Previous complaints about linear models:
Linear Lasso and Regularised
What's wrong with linear lasso validation?
Let us know if you have any questions
Cheers