Deploying ML/Deep Learning Models to Production

Re:Deploying ML/Deep Learning Models to Production by WestCoastJustin on Monday January 28, 2019 @05:54PM (#19021881)

It depends on that software you are using. But, you might want to check out https://cloud.google.com/ml-engine/. This works pretty good for TensorFlow. Basically, upload your model and it'll expose it as an API. Then you can flip through versions as you evolve it.

If you use something like flask/Docker you are totally owning the entire pipeline and that might be a good/bad thing. By own, I mean hosting it yourself. Do you really want to own the pipeline, are you getting anything out of it, is this a competitive advantage to you some how? If not, you probably want to just off-load it to something else. Then, someone else can worry about all the production issues, and you can focus on what you're good at.

[ Parent | on HN ]

Re:Deploying ML/Deep Learning Models to Production by prasann16 on Monday January 28, 2019 @06:20PM (#19022073)

I am not getting anything out of building my own pipeline. I am using Pytorch. Do you know something similiar to Google ML Engine for Pytorch where I can upload my model and it will provide an api for inference?

[ Parent | on HN ]

Re:Deploying ML/Deep Learning Models to Production by WestCoastJustin on Monday January 28, 2019 @06:25PM (#19022120)

I only know about the Google one. Looks like they have added support for scikit-learn, XGBoost, Keras, and TensorFlow (in-beta). So, you might be able to just use it by porting some of your stuff. I'd just run through the examples to see if it'll work for what you want.

[ Parent | on HN ]