Graph Visualization: Added the General Chunk algorithm, multiple stacks to address debug markers, colors for nodes and unit testing.#2550
Conversation
bjoeris
left a comment
bjoeris
left a comment
There was a problem hiding this comment.
I think some of the functions and data structures are complex enough that they are difficult to understand from the code alone, and some brief comments would probably help with clarity.
| func addDebugMarker(builder *labelForVulkanCommands, from, to int) { | ||
| level := builder.labelsInsideDebugMarkers[len(builder.labelsInsideDebugMarkers)-1].GetSize() - 1 | ||
| builder.numberOfDebugMarker++ | ||
| func (dm *debugMarker) addDebugMarker(from, to int) { |
There was a problem hiding this comment.
I still don't really understand how addDebugMarker and checkDebugMarkers are working. A few comments might help.
There was a problem hiding this comment.
I implemented a different way to handle the debug markers, I think now it is very simple to understand the idea, also I added comments to better understanding.
| } | ||
| } | ||
|
|
||
| type chunk struct { |
There was a problem hiding this comment.
I think this struct and the fields could use brief descriptions of their purpose. The types and names by themselves do not by themselves make it very clear what these variables represent or how they are used. The other structs holding somewhat complex internal data structures like this could also probably use some brief comments.
There was a problem hiding this comment.
I added a brief descriptions of structs and fields , most of them in regard to the chunk algorithm.
| } | ||
| } | ||
|
|
||
| func (g *graph) joinNodesInChunksByFrame() { |
There was a problem hiding this comment.
Is there a reason that joining nodes by frame, and making the chunks have to be done together like this? Could we have separate joinNodesByFrame() and makeChunks() steps?
There was a problem hiding this comment.
I separated them in two different functions, join joinNodesByFrame() and makeChunks(), in addition I added the config for the global variables used in the chunk algorithm.
| } | ||
| } | ||
|
|
||
| func makeChunks(nodes []*node) { |
There was a problem hiding this comment.
I think this function could use a comment describing its purpose.
There was a problem hiding this comment.
I added a comment to describe the purpose, also I renamed some variables to better understanding of the code.
ecapia
force-pushed
the
graph_visualization_6
branch
2 times, most recently
from
acf8e73 to
46bd808
Compare
| // for the range [0, 1, 2, 3, ...]. | ||
| positionToChunksID [][]int | ||
|
|
||
| // builded checks if the K-ary tree was builded for this chunk. |
There was a problem hiding this comment.
You probably want built instead of builded
There was a problem hiding this comment.
You are right, I fixed this, thank you.
AWoloszyn
left a comment
AWoloszyn
left a comment
There was a problem hiding this comment.
One tiny nit, then its good for me too.
to address debug markers, colors for nodes and unit testing. General Chunk Algorithm: This algorithm creates super-nodes to avoid display more than a fix number of nodes in every level. For every level, the nodes are grouped in super-nodes based on command names, then if the numbers of nodes exceed the maximum limit, chunks are created, and if in the new chunks the numbers of nodes exceed the maximum limit, then new chunks are created and so on. Results: Applying the General Chunk Algorithm is possible to visualize graphs containing millions of nodes and edges. The time is reduced from ~20 seconds to ~2 seconds to expand super-nodes in Tensorboard. Note1: All "op" attribute for nodes in pbtxt format must be different, since the find_similar_subgraphs algorithm from Tensorboard computes the similarity between nodes with same "op". Note2: Adding colors to the nodes increase the time for expand super-nodes.
ecapia
force-pushed
the
graph_visualization_6
branch
from
46bd808 to
1b8a72b
Compare
6 participants
No description provided.