Learning approaches can be divided into the principles of "how to teach" and directly "what to teach". Even with a good curriculum (“what to learn”), there is little end-of-life blow-up if learning is ineffective. Therefore, you first need to determine which principles are effective.
I cite only the principles that I use myself, ranging from the obvious and up to those that many know but do not use in self-education, although they are no less universal. In general, these are very general principles of learning something, which I slightly adapted to the specifics of machine learning, and indicated specific examples.
These principles helped me to move relatively quickly from 1C to a data scientist and in two years to grow to the level of a signor, in terms of salary and autonomy (link to the previous article about this)
You need to understand the essence, principles and concepts, intuition. Don't try to just remember
"Knowing some principles eliminates the need to know many facts"
"Behold at the root" (Kozma Prutkov)
At school and college, many fell victim to approaches when it was important to learn something by heart or to know some unimportant details. I remember how it struck me that many of the top American professors who teach courses on coursera.org try to first explain the intuition of different approaches, including completely mathematical ones, instead of showing the formulas. In practice, you will often not need to know the exact formulas (or you will have time to remember them). But in order to make the right decisions, it is necessary to firmly keep in mind the principles and logic that lies behind different approaches.
Therefore, when you study mathematics, the principles of different algorithms, or even individual formulas, it is important to concentrate on understanding the basic things and principles behind these formulas, and not on memorizing them.
(Andrew Ng), , , . , ( ).
, :
, - ,
( )
-
"" -
. , , ,
, :
(log-odds), ,
log-likelihood: log? ,
. , ,
" - 10% 90% ". .
, . - . .
, , - , , - , .
:
: 2-3 . , . : . - -
: -. , , Python -, http://pythontutor.ru/
: - : -, - , , - : " , ?". - ! , , / - , . ( , , , ).
2-3 . .
: ,
, , . , - - .
, . , , , . , . , , . .
, :
machine learning data science , , . , / . ( .).
, , , . . (, )
, , , . /, , . " " - , , ( ), .
3- , /, : , , , .
" ", , , , . , Andrej Karpathy, Tesla, - , .
:
, ; , ; , , .
.
- , .
, - , ! , .
- . .
", : — , — !" (.).
- . , . : , " ", .. , . . , , .. , (: Andrej Karpathy ).
- ( ), , . . , , , , . , ( ).
:
" . . . , . - . . , .
. . , , , …"
- Explore/exploit trade-off.
, , :
- . - - , , . - , , (, , ). : (explore) , (exploit). , , : - , , .
, , . , , , - - - (, ).
, - , . :
, / . , / , , - . . , , , - , . , , , , (!), , , , , , , - ..
, , .
. . , , . , - . , ? . , udemy.com coursera.org. : , - . /. , , , " ".
, / 20% . , , .
- : !
, .
, , , . . (!) , .
data science, . , , / . , " ": " -, ".
. , . , .
, , anki ankidroid . , . , ankidroid bash. , , , . .
. , , data science, . ( .. ).
.
, ( 2 ) , , .
: , .
Maybe we will talk on the phone once and this will be enough for you. But I'm here to help replace paid course programs by giving you a personalized curriculum and talking about the best books and online courses and how to find them for free. That is, you can learn not only more efficiently, but also cheaper. Your promise is enough for me that when you feel the result, you will pay for it as you see fit.
self.development.mentor in the gmail.com domain, Oleg