mpp-splitter

mpp-splitter copied to clipboard

Include proper feature engineering and pruning to create a reasonable cost function

4

A preliminary test indicates that using `1.6*log(ppm+1)+log(uncertainty_unit_cost)` as the cost function works pretty well for feature engineering. This has a nice implication on runtime. As with proper feature engineering one...

renepickhardt

As indicated in #12 the pruning can be done once we have proper feature engineering by taking a quantile of the cheapest channels. However I guess the pruning should respect...

renepickhardt

in this PR I introduce a notebook that showcases how one may do `Pickhardt Payments` (which are according to the notebook defined as: Using [probabilistic payment delivery](https://arxiv.org/abs/2103.08576) in a round...

renepickhardt

Simple graphs vs multigraphs

1

At least for maxflow scaling techniques become more efficient on simple graphs O(nm) instead of O(m^2) as for multi graphs as described in https://youtu.be/7QPI3kBIKv4 at around minute 71 mark. I...

renepickhardt

Study braess paradoxon

1

https://en.m.wikipedia.org/wiki/Braess%27s_paradox I think in lightning a similar thing can happen. Channels only allow for a certain throughput (htlcs/sec) but if a high liquidity / low fee channel is created many...

renepickhardt

Proof correctness and runtime of the roundbased algorithm. Especially payment delivery is not np hard

1

We know from theory that finding a single max likelihood flow is a mcf problem and thus polynomial. Thus the argument in a completly static network (where noone but us...

renepickhardt

Rusty mentioned yesterday that he considers giving an order of magnitue of the balance in a failing channel inside the error onion. I just realized if a node has not...

renepickhardt