How hard it is to write a neural network from scratch for example in C++? Not in terms of what to take and the library which implements all the functionality, namely zero.

I'm interested with the position to understand the complexity of it all. For example, Tensor Flow is how many man-years?

Can this now be done alone by one person or are there a few dozen ought to work on this?

He wrote nothing going, just the scale of complexity is interesting and what libraries are needed for this? To work with multidimensional arrays and all?

I'm interested with the position to understand the complexity of it all. For example, Tensor Flow is how many man-years?

Can this now be done alone by one person or are there a few dozen ought to work on this?

He wrote nothing going, just the scale of complexity is interesting and what libraries are needed for this? To work with multidimensional arrays and all?

asked June 26th 19 at 14:03

3 answers

answered on

Solution

I went about 400 rows. This included the perceptron and its "alocatel" that slipped him the picture and hit/praised depending on the outcome. Plus it was napico a C# app with the UI that is used generate weight for identification. There is nothing difficult if you know the materiel.

PS. But right now, I look at the code and don't understand why I wrote some parts of it so=)

PS. But right now, I look at the code and don't understand why I wrote some parts of it so=)

And there are some specific libs needed? To work with a thread algebra complex arrays or n-dimensional, or more banal? - paula.Ste commented on June 26th 19 at 14:08

No, only cycles. Of course my implementation and some were not with Tensor Flow, but black-and-white signs she recognized without problems. - buford_Hand41 commented on June 26th 19 at 14:11

answered on June 26th 19 at 14:07

To write if only it worked, not universal, just for your task - elementary.

To teach - hard.

Well and optimize for speed is long.

------------

If you need to appreciate the complexity of ready-made solutions -**look at the source code for** those available.

To teach - hard.

Well and optimize for speed is long.

------------

If you need to appreciate the complexity of ready-made solutions -

answered on June 26th 19 at 14:09

There's not a problem to write so that worked without errors. But to write so that

1) Worked quickly

2) it could work on the GPU

3) Or to work in a cluster

4) Was quite universal in terms of use

5) Easy to debug

6) had a convenient API

...

Here it is, I think, takes most of the time when designing things like Tensorflow.

1) Worked quickly

2) it could work on the GPU

3) Or to work in a cluster

4) Was quite universal in terms of use

5) Easy to debug

6) had a convenient API

...

Here it is, I think, takes most of the time when designing things like Tensorflow.

Find more questions by tags Neural networks