The trading bots wants access to the market structure, queue lengths really. Trader bot issues a search graph, pit boss lets it run. Finite, protected, countable, like a byte code for traders. True JSON is a candidate for this search graph. Trader bot brinds its own matching function. Make buy and sell bids right there, inside the instruction cache. Protection is key, calls in and out of join only to trusted miners.
Trade graph preview, by preview bot, will proof read the search graphs then clear them for use. The proof is automatic, an exploration of all possible exits prior to releasing the graph onto a join with market structured queues. All of this, sandbox compatible.
Neural nets
Consider a set of homomorphic, even identically shaped, directed graphs. Each of the graphs distribute results from a feature detector, distributing the results onto their homomorphic cousins. We get a do loop of do loops, all the weightings set properly in the Hopfield net. Most of AI, even the "if/then? networks, can be modeled as a join.
Took a software break and tested the MEM module in a full integration. It seems to be working. I actually did five minutes of tax work, then froze.
No comments:
Post a Comment