Agreed? 🧐

Yeah I’d agree in most cases. I’ve seen ML folks create a reasonably working model in a short amount of time and one of the managers comes in and says we need to use Keras or pytorch or some shit like that, and everyone just ducking stares at the said manager. Slippery slope all the way.

This is one of the reasons why people think ML is just magic that happens on its own. They dont understand that a human had to spend days tweaking parameters and architecture to get a good result. And thats assuming you already have the data stored sane and normalized ^^;

So on first process we see car that is going to a machine operated by human in front of computer, we want to utilize this car into square steel form cause it’s diesel.

On second graphic we eliminated human so machine turning diesel car to square steel is fully automated.

As we can see, human is replaced by 3 circles.

For people who don’t understand the pictures there are subtitles.

I love how when it comes to ML and stuff nobody talks about the poor motherfucker who writes SQL and uses pandas and numpy to clean data together for days on end, cos the data is always shit. I’ve been there. Extremely unpleasant experience fuck me I probably have PTSD

@deadlockbreaker

Were we suppose to laugh that ml is able to find the output directly and deep learning is not ?

As far as I have seen you have to normalize features even in case of deep learning. I don't think that feature extraction is completely removed. Though linearly dependent feature are automatically handled in deep learning but you Still will have extract features. Unless the input is limited to images.

Well may be I am wrong who knows ?

So much wrong. That photo. DL still requires data cleaning, feature extraction and I'm addition PARAMETER TUNING. Which is really what takes 90% of the time.

So yeah, the comment below the picture is right. DL is not a quick replacement for analytics, it's just there as an additional tool.