Check out the first Dataiku 8 Deep Dive focusing on Productivity on October 29th Read More

Online Learning of ML models

Level 1
Level 1
Online Learning of ML models
I was trying to find a way to do online learning in Dataiku.
 
An example:
I have (x,y) data coming in the system and I want to preform linear regression on it. As the data is coming from a stream (let us assume that we have partitioned the data based on time), we build the model on the first batch/partition(p1), i.e., we have slope and intercepts (m1,c1). 
 
Next, we have new data coming in, i.e., p2. We want to update our model learning with p2. Data related
to p1 is gone and we only have (m1,c1) and p2 with us and we want to compute (m2,c2). 
 
Is this achievable with Dataiku visual recipes? I looked at references and came across 'ensemble' of models. I am not sure it is the right way in the above setting. Also, checked scoring the models, but it let's us retrain the model (assuming p1 is available).  
 
Please let me know.
Thanks.
0 Kudos
0 Replies