This is a question that is relevant to any kind of development and machine learning (ML) is no exception. But at the same time, for sure, many will ask - why is this article needed, how does your ML differ so much from the standard development, according to which articles have already been written a wagon - read, analyze and choose the right path.
On the one hand, this is how it is - and there are a lot of articles to analyze and analyze. On the other hand, there is a specificity - and the staging of ML development is somewhat different from the standard one and the work goes not only (and not so much) with the code, but with the data.
But let's take everything step by step - let's go over the differences in a nutshell, and then we'll figure out if there is a place for outsourcing in ML development and what it is.
Instead of an introduction, a few words about the differences
In fact, the main specificity of ML development is that it is not the code that rules, but the data. Of course, there is also a specificity that we do not write ML algorithms, but only use (train), but this, again, is largely about data. And what do we have? That's right - data is primarily a strategic asset of an enterprise. And by and large, ML is nothing more than the process of monetizing this very strategic asset. And how many are ready to give the monetization of their asset "to the side"?
A funny analogy came to mind ... there is a famous phrase - "Data is oil of the 21st century." So if we continue this analogy, then ML is an oil refinery. And of course you can find oil producers who sell crude oil, but most of it still refines and sells the already refined product.
Plus, of course, do not forget that the data in many companies is information containing commercial / personal / medical secrets (underline or delete the necessary information) and this also imposes a number of restrictions or at least requires increased attention.
, “ , , !”. . , , ML - . , .
1 -
, ML , ? , , , - . , , - - . , , . , "" ( ).
“ ?” - . , . . , , ( ), . .
- ML . - - PoC (Proof of Concept). ( CRISP-DM):
0 - . . - .
1 - . , . - , , , .. , .
2 - . . , . .
3 - . - . ML, “ ”.
4 - . :
ML
.
- ML DSaaS (DataScience as a Service), . - - . … .
2 - -
. ML , ? - . , “ ” , . ( ), . , ML - . . , . -, , , , , :
( )
, -
ML , . , - - .
, , , . . - - . , :
, :
- , ?
- - ,
- - , , ( )
- - , , , .
, . :
- . ?
( ) , .
- - - . ?
- //
, ,
/ , “ ” “”, “ .. ”, . - ML , “ ”, " " “ ” ( , , ..) .
, - :
- , . ML - , “ ”.
- , . ML - . .
, :
ML,
- PoC , -
-