Author: Joost Jager 2021-11-15 19:26:51
Published on: 2021-11-15T19:26:51+00:00
The context of the conversation is about probability-based routing and feature engineering. The speaker mentions that reliability can and should be expressed as a probability, which is discussed in a prior article. They go on to explain that they implemented a form of probability-based routing in lnd in March 2019 and refined it through several rounds. Regarding feature engineering, there are two steps: feature selection and combining features. Feature selection involves computing a min cost flow with a chosen cost function that represents the weights on the edges of the graph. The cost function should represent a combination of fees and reliability, but there are other potential features that could be used in the future to deliver more optimal results. Combining features involves using a weighted linear sum or harmonic mean to combine selected features. The speaker wants something anchored to an understandable starting point, so they worked with payment attempt cost, which is a virtual satoshi amount that represents the cost of a failed attempt. By putting a high price on failed attempts, pathfinding will tend towards more reliable routes even if they require a higher fee. Ultimately, the fee/reliability trade-off is up to the user to decide on.
Updated on: 2023-06-03T06:29:46.707255+00:00