Edit on GitHub

Glossary

There are too many similar types (and used to be 4 more!) in the lc3 search code, so it feels like a good idea to list them.

Vocabulary

First, let’s fix some vocabulary and variable names. It’s still in flux and quite inconsistent in the code, and I’m not sure whether to use “classic-style” names like q, n, or use more descriptive names like num_visits, node_value.

short	long	Description
n	num_visits	Number of visits to a node
m	moves_left	Number of moves (actually, ply) left, as predicted by NN
w	win_probability	Probability of winning
d	draw_probability	Probability of drawing
l	loss_probability	Probability of losing
v	net_win_probability	w - l. This is the value we store in the node for compactness.1
q	node_value	q value to use in the \(Q+U\) formula. 2

Some of these values may have a prefix:

short	long	meaning
raw_	raw_	Single node value (e.g. as returned by the NN)
agg_	aggregated_	Aggregated value (e.g. average of multiple nodes)

Data types

Variation

(I consider renaming it to TreePath or NodePath).