Print the actual pruning rate

[AlexKoff88]

Add calculation of average pruning rate for the whole model (no for pruning layers) and print it after the table with per-layers statistics.

[AlexKoff88]

cc'ed @mkaglins

[AlexKoff88]

We also need to revise information that is printed after each epoch. I found some strange numbers that make it unclear. Just check out:

+---------------+---------------+------------+-------------+-------+-----------+
| Name | Weight's | Mask Shape | Mask zero % | PR | Filter PR |
| | Shape | | | | |
+===============+===============+============+=============+=======+===========+
+---------------+---------------+------------+-------------+-------+-----------+
| MobileNetV2/S | [64, 192, 1, | [64] | 12.500 | 0 | 0 |
| equential[fea | 1] | | | | |
| tures]/Invert | | | | | |
| edResidual[7] | | | | | |
| /Sequential[c | | | | | |
| onv]/NNCFConv | | | | | |
| 2d[2] | | | | | |
+---------------+---------------+------------+-------------+-------+-----------+
| MobileNetV2/S | [64, 384, 1, | [64] | 12.500 | 0 | 0 |
| equential[fea | 1] | | | | |
| tures]/Invert | | | | | |
| edResidual[8] | | | | | |
| /Sequential[c | | | | | |
| onv]/NNCFConv | | | | | |
| 2d[2] | | | | | |
+---------------+---------------+------------+-------------+-------+-----------+
| MobileNetV2/S | [64, 384, 1, | [64] | 12.500 | 0 | 0 |
| equential[fea | 1] | | | | |
| tures]/Invert | | | | | |
| edResidual[9] | | | | | |
| /Sequential[c | | | | | |
| onv]/NNCFConv | | | | | |
| 2d[2] | | | | | |
+---------------+---------------+------------+-------------+-------+-----------+
| MobileNetV2/S | [64, 384, 1, | [64] | 12.500 | 0 | 0 |
| equential[fea | 1] | | | | |
| tures]/Invert | | | | | |
| edResidual[10 | | | | | |
| ]/Sequential[ | | | | | |
| conv]/NNCFCon | | | | | |
| v2d[2] | | | | | |
+---------------+---------------+------------+-------------+-------+-----------+
| MobileNetV2/S | [96, 384, 1, | [96] | 16.667 | 0.083 | 0.083 |
| equential[fea | 1] | | | | |
| tures]/Invert | | | | | |
| edResidual[11 | | | | | |
| ]/Sequential[ | | | | | |
| conv]/NNCFCon | | | | | |
| v2d[2] | | | | | |
+---------------+---------------+------------+-------------+-------+-----------+

I don't understand why, for example, "Mask zero" is 12.5%, while "PR" and "Filter PR" are 0. @mkaglins to comment.

[mkaglins]

@AlexKoff88 That means weights aren't zeroed for some reason (or became non-zero).
This is strange since applying masks and stats calculation are near in the pipeline (I will re-check it).

[AlexKoff88]

Maybe this issue is related to #353. Anyway, we need to change the column names and even the set of columns keeping only something like "Expected PR", "Actual PR".

[mkaglins]

@AlexKoff88 you suggest delete Filter PR or just rename it too?

[AlexanderDokuchaev]

Will added actual pruning rate in flops

[AlexanderDokuchaev]

#367