If we want to add figure for SD size and attractor scaling vs depth for NK models, we need the following:
[ ] Get SD size and depth for random NK up to depth 9, using full expansion without attractor search. Depth 9 is for easy comparison with bbm-inputs-random. -> [size vs depth figure]
[ ] Get the number of attractors and depth for random NK up to depth 9, using source SCC expansion with attractor search. -> [n_att vs depth figure], [checking existence of motif avoidant attractors]
Note that bbm have the source nodes fixed, whereas the random models retain the source nodes. Fixing the source nodes of random models from the start could allow more accurate comparison. Let me know what you think about this.
Attractor scaling
If we want to reproduce attractor scaling for NK models in the Science Advances paper with K=3, we need below:
[ ] Get the number of attractors for random NK, aiming for N > 4096, using source SCC expansion with attractor search. -> [N vs n_att figure]
The current results we have for N <320 is not enough to get the scaling. Also 100 samples for each network size is not enough.
Benchmarks
I'm not very sure about how this should be done, so I make note of what was done in pystablemotifs
[ ] attractor search: NK models, N 10~100, 10 models each per size, 12 hour cutoff
[ ] attractor search: empirical models, 5 models with N ranging from 40 to 101, 12 hour cutoff
[ ] control
I'll first try running them on my pc. If it seems impossible, I'll ask @jcrozum to run them in a stronger machine.
Succession diagram scaling
If we want to add figure for SD size and attractor scaling vs depth for NK models, we need the following:
Note that bbm have the source nodes fixed, whereas the random models retain the source nodes. Fixing the source nodes of random models from the start could allow more accurate comparison. Let me know what you think about this.
Attractor scaling
If we want to reproduce attractor scaling for NK models in the Science Advances paper with K=3, we need below:
The current results we have for N <320 is not enough to get the scaling. Also 100 samples for each network size is not enough.
Benchmarks
I'm not very sure about how this should be done, so I make note of what was done in pystablemotifs
I'll first try running them on my pc. If it seems impossible, I'll ask @jcrozum to run them in a stronger machine.