Machine learning · Python

Big Tech Machine learning frameworks vs Saas based frameworks

Herbert Roy George Enterprise and Technology expert with enormous digital experience

May 2nd, 2019

Hi all

I understand that machine learning models take a while to build .

I'd like to understand if using a service like Google ML or IBM watson or Microsoft's AI reduces the time or effort taken to build a proper ML product ? Do they come with easily customizable pre built models or is it too straight jacketed ? How easy is it to train those systems or would you advice building a custom ML system from scratch using some Iaas / Paas?

Herbert Roy George Enterprise and Technology expert with enormous digital experience

May 2nd, 2019

Irrespective of what my objectives are wouldn't it be possible to gauge an out of the box ML system/service vs setting up one from scratch ?

Mukesh Mithrakumar Machine Learning Engineer

May 2nd, 2019

It depends on how big you would like to scale in the future and if you are willing to be tied up with Google or IBM but the short answer is, it takes less effort to build using Google ML or others, but you still need to know about ML to a certain extent to really tune your product and get the best performance. And they do come up with pre built and pre-trained models. And you dont have to worry much about training since it's already pre-trained, the amount of training you will have to do for your dataset will be minimum but my advice, if you are looking for a short term product, Google ML will be good but if you are looking for a long term solution and which always require you to innovate, its better to spend the initial time building your custom ML system

Louis Lim Founder, CEO Helios Eos

May 2nd, 2019

I doubt anyone can give you an answer without knowing anything on your objective. But you can try using Saas, it is not really that expensive, unless if your requirement and objective is really specific. To understand more, you should consult with a few specialist to understand what architecture is required and other feasibility studies. At the same time, you can really understand who can do the job better. You can connect to me, if you like me to share my experiences building a framework.

Sean Reynolds Principal Robotics Engineer

June 13th, 2021

Part of the problem with google ml in the cloud is that you have to use it in their environment. It’s not like they are just wrapping tensorflow and giving you a trained tensorflow model at the end. Cloud based ml solutions are best for proof of concepts but any ml engineer will have their own environment already set up and could build you a model with a given data set pretty quickly.

(my casual two cents)