How can I start&elaborate with Machine Learning? I don't

https://www.reddit.com/r/MachineLearning/

>>61271881
Unfortunately it's too late.

Successful machine learning researchers are identified in elementary school machine learning competitions. Only the most creative, innovative, and gifted students are selected. If you were never aware of the process, then it means that you failed in the secret initial qualifiers, and weren't even close to earning a place in the program.

This process may sound harsh, but it would simply be cruel to try to train someone in the art of machine learning if they don't possess the raw talent.

>>61271899
Thanks!
Reddit seems helpful if not related to humor or politics.

>>61271904
btw since there are only a few machine learning experts in my country, Maybe I have a chance to ride a train.

Bp

File: r.png (43KB, 824x231px) Image search: [Google]

43KB, 824x231px

What do they mean by objective?

Trying to implement this, have seen nobody else use that term.

>>61274099
it is the loss function, did you skip some chapters on your material?

>>61271881
>tfw want to learn machine learning but don't know where to start, most only tutorials are the type of tensor.fit() or scikit.fit() tier shit where you don't learn the stuff behind meaning you won't be able to implement your own shit outside that tutorial.
halp?

>>61274608
I see. Thank you. I just wasn't sure what they were referring to.

I'm learning from random resources I find online. Seems people use various terms to mean the same thing, and I haven't seen this one used before.


So if I'm understanding this correctly, should I increment the final error value by 0.5*λ*w^2 of each and every weight in the net? Even with a regularization strength of, say, 0.0005, that could get out of hand pretty quick, surely?

>>61274685
Here

http://karpathy.github.io/neuralnets/

Extremely simple way to explain neural networks, and you'll have full capacity to implement one yourself after this (and it will be slow as fuck, but you can't really expect anything else).

>>61274685
go through cs231n lectures

>>61274734

 >>> w = tf.to_float(tf.reshape(tf.range(0,6), shape=(2,3)))
>>> loss = tf.nn.l2_loss(w)
>>> sess.run([w, loss])
[array([[ 0.,  1.,  2.],
       [ 3.,  4.,  5.]], dtype=float32), 27.5] 

>>61275191
Nice.

Well, I'm trying to learn the stuff that happens behind the scenes