.NET?

[minor surge]

How much would I complicate my project if I decide to do the AI part of it in .NET? ( AI part = detecting certain objects and returning True if they are detected)

[minor surge]

.NET?

[ripe egret]

you have to create AI system using Yolo or OpenCV or CNN and integrate that with net. not complex.

[minor surge]

And what about:

Training a model to detect cookies.
If a cookie is detected, return True

I used Yolo and OpenCV in python

🤔

[ripe egret]

what do you mean ".NET"?

[minor surge]

.NET the C# framework

& ML.NET

[main glade]

guy wants to train in .NET itself

unadvisable

do what @ripe egret suggests, train in python, export to onnx whic can hopefully work in .net

[topaz schooner]

!BeastBlaze

[minor surge]

They claim to get better accuracy and more speed

[main glade]

accuracy sounds like bs

speed maybe

[minor surge]

It’s for same dataset

[main glade]

they dont even include pytorch

not to mention scikit learn can vary in accuracy based on what model u use

but anyways

[minor surge]

I saw something like it runs through all the algorithms for training and decides which gets best results

[main glade]

tbf i think its neither faster nor more accurate

cuz fastest is usually tensorRT

opencv dnn afaik is a tad bit faster than TensorRT even

[main glade]

they used it for both scikit learn and their mlnet?

[minor surge]

Also it has a nice user interface

[main glade]

go ahead and use it ig

[minor surge]

[main glade]

i wont recommend it but if u are so obsessed go ahead

[main glade]

?

[minor surge]

One sec

[minor surge]

What?

[main glade]

the "goes throw all algos and finds best"

[minor surge]

That’s what ml.net does

[main glade]

did they do that for both sklearn and mlnet?

this selection process?

[minor surge]

One sec I’ll find the video

[main glade]

so they used the horrible selection process of their own framework but not the much more superior auto-sklearn framework that makes weighted ensembles..

u can believe the bs advertising if u like

and be obsessed with it if u like

[minor surge]

One sec

[main glade]

but just consider that everyone uses pytorch/python (and btw sklearn is likely not even the best for ur task)

[minor surge]

The graph is from.net official website not the video

[main glade]

so it has a lot more support

[minor surge]

I’m not obsessed with it

[main glade]

u sound like u are

i am obsessed with python for ML/DL dev but for good reason

(most people are too, even if they dislike python for whatever reason)

[minor surge]

I’m working on a project that isn’t very innovative so I want to make up for that by using/researching less used AI/ML tools

[main glade]

sounds like a not very innovative idea

if u can even call it that

[main glade]

just like mlnets poor selection strategy

[minor surge]

I might figure out something useful

[main glade]

doubtful

but anyways yeah do whatever u like if u dont wanna take advice from people

[minor surge]

We’ll debate in 6 months dw!

[main glade]

good thing i am not your advisor and its probs some chump who doesnt know anything about ml

[main glade]

doubtful

[minor surge]

[main glade]

they might actually give u a passable grade

for doing useless/stupid stuff

[minor surge]

AI is a small part of the project

[main glade]

yeah go ahead do whatever

[minor surge]

Ty for helps

[main glade]

its like if i used a bad model from sklearn ofc the best model from mlnet will outperform

what dataset even is that?

im gonna give u about 1-2 years of ml advice in one page, use this: https://automl.github.io/auto-sklearn/master/

well actually dont cuz what u are doing is a vision task, so use pytorch

but anyways in the end its your choice

[main glade]

this actually uses a decent non-dumb model selection strategy (unlike mlnet)

[minor surge]

I’m googling a bit and the only thing that people are complaining about when it comes to ml.net is that there’s a small community and not many resources

Which makes sense because it’s not the industry standard

[minor surge]

“There are several ML NET benefits. It has a built-in AutoML allowing you to avoid choosing the best type of model manually.” The thing I mentioned earlier

[main glade]

bro idc i have given u my advice if u dont wanna take it, be my guest