ML, Deep learning, CV, and NLP Projects with code
github.comLists can't be Show HNs, so I've taken that out of the title.
https://news.ycombinator.com/showhn.html
Generally speaking, they don't make great regular submissions either, since HN is already itself a list, and a pointer to a pointer to a pointer is too much indirection.
https://hn.algolia.com/?dateRange=all&page=0&prefix=true&sor...
Bummer, only 500. Anybody have a link to a repo of 1000 random Medium articles with AI/ML?
This is very cool, but the level of indirection is strange. It's a repo, that contains links to medium posts, that aggregate posts from thecleverprogrammer. And it all seems to be owned by the same person. Why not make just one aggregated list?
If you like this, I do a weekly roundup of open source projects that includes an interview with one of the devs you can subscribe to.
Quantity vs Quality, I guess. I randomly selected a link and ended up in this article about NER. It explains how to train a named entity recognition model but skips some functions and in the end where it says "testing the model" it shows results of a pretrained spacy model wtf?
[0] https://thecleverprogrammer.com/2020/08/04/named-entity-reco...
I'm not sure about this specific post but I've noticed an increasing number of people who sort of borrow (steal) from a few different sources and repackage it as their own Frankenstein's monster.
There is a lot of noise. A lot of data virgins seem to do it for audience building/"enthusiasting"/Tweeting/YouTubing/Mediuming/motivating/inspiring/what's holding you back"ing"/ML in production.
Similar to people "teaching" martial arts/"self defense"/Close Quarters Combat: "And your attacker will hit you with his right hand, and you'll grab his pinky, and step on his toes to hurt him. Repeat until he says owie!".
It appears that most of the content about the deed is produced by people who never did the deed. That content theng goes through human centipedes.
However, I think that it is a spectrum. If all content about disciplines was only extremely technical and legitimate, it wouldn't spill over and reach people on the outside.
The important thing is that non-technical people can understand it.
Like hiring managers.
Nice sarcasm you got there. Just to unsarcastify this: it would be really bad for hiring managers to be clueless. They'd be hiring for all the buzzwords with the wrong expectations, have meaningless job postings.. Wait a minute...
I was thinking about children being able to get thrilled and become interested and get in the field.
Anyone have recommendations for examples of deploying a basic ML image classifier as a restful api? or examples of a multi-attribute image classifier?
It's not really that different from deploying any other sort of RESTful API.
If you want a framework you could use Seldon but could probably just as easily write your own with Flask/fastapi.
Wow, FastAPI looks like a very cool framework. I guess I was moreso curious in how one would take a model that works in a Jupyter notebook and restructure it to be callable / have a self re-training data pool of images that were submitted. (obv this feature would pose some legal risk)
Convert the notebook into a python script and use that with FastAPI