I used to hate that at the beginning but Java's strongly-typed data object strategy is paying off HUGELY.
Once I started writing critical programs, in tight schedule, with all sorts of coders involved, it makes so much sense.
I remember my first "aha" moment. When I was writing my first really complex program (I mean, it wasn't super complex but it had to be done from scratch, as there was nothing like it at the time): So, I was coding and noticed how the language steers me towards creating representations of the real stuff that my program handles.
I really wanted to keep everything as strings jumping between static methods... But kept reaching these situations where it's so much redundant code everywhere that I can just define a type and handle things "in one place"... But then when I started putting THIS type together, it didn't make sense without THAT type too... So, I noticed about half of my time budget went on just creating types and methods that nothing is yet using... I got scared and though I'd have to stay after work to actually put things together into an actual solution!
But... The moment I lined all my types in a row... I was like... Wait, all I need is this little trigger here and things will just work... But that can't be true, can it?
I tried, it worked and we are using this project, with those initial types and nothing new, for two years now
And now I actually like how Java tries to enforce people to represent reality in object types
Although that is not necessarily a Java "feature" - I bet C# is very much that way (although I don't know, I never really programmed in C#)...
Then again, nothing about TensorFlow is inherently "Python". It could have been a Java library just as well (I think?)
oh, and (again, not a feature) the fact that pretty much everything is implemented in C is kinda awesome for C :)
Honestly, I don't think any of those new languages have a *Language* feature that makes them super special. It's all a popularity contest and, well, no one is going to start writing a critical library in multiple language... They are going to choose a language and then flesh out the library... And the chosen language becomes the "favorite" of people using that library, because the library is popular, not so much the language.
Actually, I thought that Prolog would have been the language of choice for all the ML going on right now with Python.
Prolog actually forces you to think about branching out possible outcomes and then backtracking to the one that is "true". I did quite a bit of work back then with Prolog which I think now is pretty similar to ML.
Guess the ML guys just wanted to do it in Python and the rest is history...
Once I started writing critical programs, in tight schedule, with all sorts of coders involved, it makes so much sense.
I remember my first "aha" moment. When I was writing my first really complex program (I mean, it wasn't super complex but it had to be done from scratch, as there was nothing like it at the time): So, I was coding and noticed how the language steers me towards creating representations of the real stuff that my program handles.
I really wanted to keep everything as strings jumping between static methods... But kept reaching these situations where it's so much redundant code everywhere that I can just define a type and handle things "in one place"... But then when I started putting THIS type together, it didn't make sense without THAT type too... So, I noticed about half of my time budget went on just creating types and methods that nothing is yet using... I got scared and though I'd have to stay after work to actually put things together into an actual solution!
But... The moment I lined all my types in a row... I was like... Wait, all I need is this little trigger here and things will just work... But that can't be true, can it?
I tried, it worked and we are using this project, with those initial types and nothing new, for two years now
And now I actually like how Java tries to enforce people to represent reality in object types
Although that is not necessarily a Java "feature" - I bet C# is very much that way (although I don't know, I never really programmed in C#)...
Then again, nothing about TensorFlow is inherently "Python". It could have been a Java library just as well (I think?)
oh, and (again, not a feature) the fact that pretty much everything is implemented in C is kinda awesome for C :)
Honestly, I don't think any of those new languages have a *Language* feature that makes them super special. It's all a popularity contest and, well, no one is going to start writing a critical library in multiple language... They are going to choose a language and then flesh out the library... And the chosen language becomes the "favorite" of people using that library, because the library is popular, not so much the language.
Actually, I thought that Prolog would have been the language of choice for all the ML going on right now with Python.
Prolog actually forces you to think about branching out possible outcomes and then backtracking to the one that is "true". I did quite a bit of work back then with Prolog which I think now is pretty similar to ML.
Guess the ML guys just wanted to do it in Python and the rest is history...