Which is the suggested service in Google Cloud to store Tensorflow models for predictions?
Faraz Khan
I've been reading about the different possibilities since morning am very confused about the best possible way to deploy my tensorflow model to google cloud.
I have a locally trained tensorflow model that I want to call from within my flutter app. From what I understand, there are a number of ways to achieve this:
- Deploy my model using the app engine and use REST API from within flutter to call the stored model.
This seems attractive to me as the examples I read allow me to make use of python libraries and write the predict function in python. Using this service, the app can be called via REST API from within flutter. But this seems unnecessary as I will not be using the web app launched, just need the predictions
- Store my model using ml-engine/ai-platform and use firebase functions to call that model.
I see this as something very similar to the above point as the model will be stored in google cloud ready for prediction but without the web URL or web app. This is more near to what I need as the web URL will not be needed to access the predictions, but I cant use python to call the model? The video I saw on this used Typescript to call the model stored in google cloud, but that video was from 2018 and it may not be the case anymore.
- Tensorflow serving using Docker
I only came by this while searching for solutions, but this appears to be a solution for someone hoping to host models on their local machine and make that available to people having access to that machine, so this is not what I want, but just added here for completion. [please correct me if my understanding is wrong]
I am quite confused between the above options, which one should I go for? Is one more expensive than the other?
There is so much material to read online when it comes to the above topic that I am kind of lost now and would appreciate a more experienced/expert opinion on this to give some guidance. I don't want to deploy something only to find out it was not the suggested solution to my problem. Any guidance would be appreciated.
Thanks
