How do I get into artificial neural networks and machine learning as a hobbyist? I have above beginner-level python skills.
>inb4 python is for plebs
>inb4 some amateur is trying to tackle one of the most complex fields of computer science
I'm just curious bruh.
It's actually not that hard to understand and implement. You just won't be able to do crazy neural networks.
>>9044686
>as a hobbyist
you learn some toolkit.
don't waste your time trying to understand any of this shit or trying to invent something new since you need stats 1 & 2, calc 1,2 & 3, and diff equations as a minimum prereq for ANNs.
good luck.
>>9044696
I have a physics degree and I already took all of those classes.
I'm just looking for a crash course that doesn't start with "this is how you make lists in python..."
>>9044699
>crash course
start taking some CS.
>>9044686
It's really not that hard to understand and implement. You can get a mathematical understanding of regression, neural networks, support vector machines, kernel methods, nearest neighbors, k-means and stochastic gradient descent with basic calculus and linear algebra. Start with the Coursera course.
>>9044686
YEAH WELL YOU'RE AN AMATEUR YOU DON'T GET TO BE CURIOUS AND KNOW. YOU DON'T GET TO KNOW.
YOU'RE NOTHING, FUCKHEAD. NOTHING.
>>9044686
Watch Dan Schiffman's Videos. He is an excellent teacher, and he doesn't get that crazy. He will also introduce you to processing, which is babby's first java. it's essential that you learn something other than python for neural networks - processing is a good place to start.
https://www.youtube.com/watch?v=XJ7HLz9VYz0
The rest of his videos are excellent too.
Enjoy!
DESU keras lol
Import keras
Dense(50)
Dense(10)
Dense(50)
Model.fit()
There u go
>>9044764
what a faggot
>>9045010
He's gonna be my husband, and you don't get to say anything about it.
>>9044699
>Crash course
Here you go, just need linear algebra/calc background:
https://functionalcs.github.io/curriculum/#orge7cc8b4
>>9044686
pretty ez
3 steps
propagate-a bunch of for loops that fill up output matrices/derivative outputs
backpropagate-
fill up an error matrix by doing those for loops in reverse and instead fixing up an error matrix
update-
using your initial input and error to update the weights
and tada! you can into NNs now
3 steps
everything else just expands on this
Siraj Raval and his hour long lectures are my favorite YouTube videos for neural networks.
https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A
>>9046891
There's also the free deep learning book
http://www.deeplearningbook.org/
>>9044696
>calc 2. 3 diff eq
For the most basic ones, high school stats and the first half of calculus will do
If you have can google the derivative for activation functions, then you really just need high school stats
Again, this is for feed forward-an ANN...just saying the barrier isn't some place that is unreachable/esoteric for a high school grad who knows python
>>9044753
Think you only need algebra to get nearest neighbors in its simpler forms
>>9046919
Also would like to mention, sure you have matrices-doesn't mean you are necessarily doing linear algebra(inb4 muh vectors)
So lets stop acting like its necessary to create a simple NN
most likely OP as a beginner would be doing dynamic programming-it doesn't have to involve matrix multiplication. You are going down each layer one by one then coming back up, not doing it at once.
>>9044686
https://www.coursera.org/learn/machine-learning