Source Code For New Apps' Tuning

Sample shader graphs

• The sample graphs of disassembly shaders can be found here [sample graph edgelist] [sample graph nodelist]

Scripts explanation

All the scripts are written in python 3.8. To run the script, please lunch a python tools like Anaconda or directly run "python xx.py"

1. Frequent subgraphs mining [code]

This makes use of python library from the link above with min_support = 25% of input graph # and min_node=5



2. Convert results from gSpan to graph edge list and node list [code]

The input to this script is the gspan output file, the outputs are the subgraphs represented by edgelist and nodelist in separate folders.



3. Measure pair-wise similarity value between graphs in each scene dataset [code]

The inputs are all shader graphs, output are the WL kernel similarity values for each pair of shader graphs



4. Select the highest similarity one from pair-wise measurement and form dataset for clustering [code]

The inputs are all the pair-wise similarities, the output is the maximum similarities between the game shader graph and across all other game shader graphs. The data is arranged in grid and will be used for training.



5. Cluster games into groups using K-Means with PCA model. The result is viewed in a 3D plot [code]

The input is the prepared dataset in previous step. Output is the train/test result with 3D plot for visualization

Useful Link

1. gSpan algorithm [code]