Building Real time Learning models
Where can I learn about Adaptive real time Machine Learning?
artml has a research paper with detailed descriptions on all models and techniques
How should we preprocess the data before creating Basic Element table?
- Convert all categorical data to numerical by one hot encoding
- Even for bipartite data (Churn or NotChurn) two separate features should be created
- Numerical features can be binned if needed to create extra features
- Feature engineering plays a crucial role for better predictions, hence non linearity can be introduced by creating new features
- For simpler cases, avoid missing values. But if needed complex techniques can be adopted based on domain knowledge to tackle missing data
How can we use complex models like decision trees for real time learning?
Currently, artml only supports linear models. But it is learned that in most of the cases more data usually beats complex algorithms. Hence we can use more data and real time feature selection techniques for making better predictions using artml. Also, create more features that introduces non linearity to build powerful models.
What is the difference between traditional Incremental/Online learning methods and artml?
All other existing methods only offer Incremental learning. Given any new data, we can use Incremental learning approach to update the linear model. But we cannot remove the effect of some old data or we cannot add or delete features in a model using current methods. artml on the other hand offers all these flexibilities (Incremental learning/ Decremental learning/ Add or delete features/Parallel & distributed processing). for further detailed info about artml, please visit artml blog.
How can we deploy artml models in a better way?
Deploying artml algorithms is similar like sklearn models. Further deployed techniques including parallel & distributed processing techniques will be discussed in the later versions of artml.
How can I get in touch?
If you have a question about how to use artml, Please refer to addtional resources for further help. For bug reports and feature requests, please go to GitHub. For everything else please email sundeeppothula@gmail.com.