On-Device training for RandomForestClassifier or Recommendation Model

General Discussion
#1

Hi I’m new to tensorflow and I’m trying to make RandomForestClassifier or Recommendation model to be enable to re-train on-device(Android) with tabular data (mostly customer interaction). Can we re-train the model on device once deployed with this new data set.

I’m referencing these examples

  1. decision-forests/intermediate_colab.ipynb at main · tensorflow/decision-forests · GitHub
  2. Adaptive Framework for On-device Recommendation — The TensorFlow Blog
  3. On-Device Training with TensorFlow Lite

But not able to understand how can I enable on-device training for tabular data on the above examples.

#2

Hi Bodhidipta_Bhattacha, welcome to the TF Forum!

As of today, TFDF doesn’t run natively on-device (converted to TFLite). It’s in our roadmap but not for the short term

cheers

#3

Thanks for reply. If in case I use recommendation model can I re-train or update it on device(Android)?

#4

I think it really depends on whether the tf ops used in your model is also available for tf lite TensorFlow 연산자 선택  |  TensorFlow Lite
I also just started to retrain on Android. Would be grateful to have a list of fully supported tf ops in tf lite for mobile onDevice training.

2 Likes
#5

As Yingding said, that dependes on the ops you need and if they are trainable on-device

more information here: On-Device Training with TensorFlow Lite

1 Like