Machine learning Academia, is very different from how its applied in a real world setting. The number of people who can launch a model and deal with the intricacies of serving a model in production are still very few. Resources like these are essential for people not familiar with ML. We need more people who can apply successfully at scale.
A top down approach to learning ML in my opinion is more efficient than a bottom's up approach, that most people try to advocate
This View point is an engineering driven view point to solve the featurization problem, the same problem can be approached in a much more simplified way which is more compatible with how practitioners work.
Have a look at some of the abstractions sagemaker has in place for A modeling workflow, abstracting featurizarion at that level is more beneficial than approaching the problem this via a core engineering driven mechanism
Based upon my experience on building a system like this, the percentage of Features reuses and searched across models is Generally lot smaller. A system which provides A publishing And a fast simple serving mechanism generally meets all the needs.
Metadata management, history, audit, lineage, search are all good to have but not critical requirements to most practitioners
Eg computing an aggregate lookup over a time range in spark, archive in S3, wrap it in a fast lookup implementation in a sickit learn transformer, and have the transformer pickle the lookup will give you offline , online parity out of box
It all depends on definition of ML engineer. Low latency high tps model serving as others have described is an interesting problem in its own right, so is management of thousands of models, add repeatability and lineage tracking to the mix and you have significant engineering challenges - |via http://bit.ly/hnbuzz01 |
I recently launched http://mistykx.com . My app allows one, to create their own photo style and make it available for free or charge a small premium for it . I have people downloading it but no has created a style yet. Still not sure if the idea might work !
I spent 5 months creating and refining the app . This involved the app, as well as the backend infrastructure on AWS . I spent a month on polishing the app and advertising the app on fb, App Store , Google
I think I need to learn to advertise better. May be add more stickiness & virality to the app
I use HNBuzz for iOS it already supports this feature . Follow a comment and receive notification when some one replies - |via http://bit.ly/hnbuzz01 |
As per the paper , you can choose arbitrary distributions , construct a fluent graph , run Monte Carlo simulation and get the result - |via http://bit.ly/hnbuzz01 |
Once I fix the iOS client I will try working on the android client ( need to learning android programming for that ) . This was my first stab at a project using objective c . For now I will be closing the Beta , but thanks everyone for signing up , appreciate the feedback . Have fixed a lot of issues , added new features , will be launching soon on the AppStore - |via http://bit.ly/hnbuzz01 |
Thanks will look into the font . Using Hacker news api to fetch everything , but they do not support batch mode , so have to make multiple Api calls , I think scraping would have been faster
. Will add support for my threads view
Appreciate the feedback , I will work on fixing some of these issues . Some of things were made specific based on user feedback but you have raised some valid points
It's more complicated than what you are saying , what you witnessed is generally called behavioral targeting which is user specific generally . Sponsored search generally is a solution which shows ads which have your keywords in it . Behavioral targeting ideally should show you stuff you might interested in buying or buy on a regular basis , if it's showing stuff that you already bought , it probably indicates a stale feedback loop between two systems