Incremental Learning Algorithms


#1

Hi All,

How to do incremental learning while building a model. Do we have a specific framework on this?

Ex: Let us say i have built a classification model on 50 GB of data. The model is good and we are using it. Now tomorrow if 1 gb new data comes in, how to incrementally train the model.i dont want to run the entire data to be passed again as it is computationally expensive . how to do this?

Any language/tool will do? i am not referring to scoring but adding incremental data and training the model in quick time.

Thanks,
Rajaram


#2

There are and they are called online learning. SVM, Baise, KNN and i think even NN has options to do incremental learning.

Matlab, tools on C/C++ have some options.

http://www.isn.ucsd.edu/papers/nips00_inc.pdf

Hope this is what you are looking after.

The other option i can think of is rebuilding the model with more focus towards newer data by readjusting your performance and observation window.


#3

Thanks Vivek for the response. Whether incremental learning can be done using tools like R, python,Spark ML or mahout? if so can you share some links for the same.

Thanks,
Rajaram.