Any good books or examples on creating an updatable model?

As I'm currently in need of creating an updatable classifier, but I have yet to find any examples.


Been reading Machine Learning by Tutorials and a little of Core ML Survival Guide. Both nice books, but written before we were given the opportunity to update our models on devie in iOS 13.


I was using an ML Classifier and updating the model in the cloud with gather data on a backend server. I would like to do the same on device.


I've got a number of columns all text and a decisiion as to what group the entry belongs to. Very much like the latest Create ML classifier project. But the model generated as of beta 5 is not updatable.


I'm hoping there is a generic solution to solve this much like the easy of creating the ML Classifier. Any help would be greatly appreciated. Web site examples, books, etc. I not against creating this in python if I could see a simple example.


Thanks,

Rob

Replies

Not sure I can bias wanting dead tree version docs w/any beta. The paint isn't dry on iOS 13 yet, so the onus is on the dev to have both realistic expectations, and the skills to bootstrap without asking what aisle the processed meats are on 😉


Start by using theiOS 13 beta forums, then take a bath in the2019 WWDC videos.


Good luck.

I'm not sure exactly what kind of model the Create ML classifier is, but it's probably a logistic regression or something similar.


It doesn't look like you will be able to make that kind of model updatable with Core ML any time soon. For iOS 13, only neural networks and k-NN can be made updatable.


There are two options that I see:


1. Convert your model into a simple neural network, and make that updatable.


2. Implement the algorithm yourself, or use one of the many open source projects from GitHub. Here's one I wrote https: //github.com/hollance/Logistic-Regression-Swift but there are many others.

Thanks very much for your help. I am really looking forward to more updates to your On device training with Core ML series.