| General Stats | |||||||||||||||||||
| Notice that including loose attractors stats change considerably… | |||||||||||||||||||
| Megabug caused previous results of #attr of DARBNs and DGARBNs to be flawed… (sometimes opposite...): some properties of CRBN are in between ARBN and DARBN, and not DARBN in between CRBN and ARBN… | |||||||||||||||||||
| What about Stats for ContextualRBNs???? Still smooth transition??? | |||||||||||||||||||
| Different results for number of attractors: frist increase, then decrease but not much… | |||||||||||||||||||
| SHOULD REDO RBNContextual graphs!!! | |||||||||||||||||||
| What about equivalence of attractors??? (same cycle attractor, but fall into the same state at different time periods) | |||||||||||||||||||
| check k=3… | |||||||||||||||||||
| mention JASSS on Model-to-Model: Disadvantage of computer modelling, but no other way of studying these things… need to double check, just like any theory… | |||||||||||||||||||
| RBNsWperiods increase exponentially attr. length with maxP | |||||||||||||||||||
| non-determ have less, but larger attractors | |||||||||||||||||||
| On Non-deterministic Updating | |||||||||||||||||||
| GARBN n coin flips per time step, ARBN one coin flip per time step, MxRBN one coin flip per PurePer time steps | |||||||||||||||||||
| no big difference between ARBN and MxRBN in stability, but big in %statesInAttr (order for free in MxRBN) (but note number of possible states is much larger for MxRBN) | |||||||||||||||||||
| algorithm too memory-expensive for large nets… | |||||||||||||||||||
| very large variance, normal | |||||||||||||||||||
| when not all states, different results (many people work with big nets, but not exhaust all possible initial states, so they get "representative" values, but some of these can vary a lot) | |||||||||||||||||||
| we can argue that natural systems do not explore all possible initial states, and therefore do not reach all possible attractors, but that is not the point. There is a complexity reduction anyway. | |||||||||||||||||||
| number of attractors increase seems to be linear, not sqrt(n). But we are exhausting all possible initial states, and not all possible nets… that could be a different result… | |||||||||||||||||||
| the smaller delta (or any order parameter), the larger the complexity reduction (low % of states in attr, shorter lengths of attr. Basins (Wuenshce)) | |||||||||||||||||||
| ordered dynamics reduce complexity too much, not flexible nor evolvable -> balance, edge of chaos | |||||||||||||||||||
| the less connections, the more complexity reduction | |||||||||||||||||||