Dear all.
I am developing an online tool for automatic feature engineering (selection and construction).
The objective is to generate new high-quality features to be used with multiple linear regression or logistic regression. It searches for equations that combine the variables of the original dataset, while minimizing AIC. Right now it uses only +, -, /, and *, and there is not internal constants.
The tool can be accessed at http://kp.unifesp.br:9000
It is running on a simple dual-core virtual machine, so the performance is not great. But it is a functional demo version.
Results are usually good, but I am certain that there is a lot possible improvements. Suggestions and bug reports are welcome!
Best regards,
Vinícius.
Bipropagation could be great. It is fast, simple, accurate and reliable.
Chapter Optimization for multi layer perceptron: Without the gradient
Dear Bojan Ploj. Thank you for your suggestion. Could you please explain how can I apply Bipropagation to construct features?
The most difficult part of feature engineering is to avoid overwriting. There is some strategies that can help you with that.
- Badges
- Science topic