preparation for mccfr abstractor + renames

Author: krukah

src/clustering/abstractor.rs

@@ -1,10 +1,8 @@
 use crate::cards::isomorphism::Isomorphism;
 use crate::cards::observation::Observation;
-// use crate::cards::observation::Observation as Equivalence;
 use crate::cards::street::Street;
 use crate::clustering::abstraction::Abstraction;
 use crate::clustering::histogram::Histogram;
-// use crate::clustering::progress::Progress;
 use std::collections::BTreeMap;
 
 /// this is the output of the clustering module
@@ -22,6 +20,9 @@ impl Abstractor {
         let mut map = BTreeMap::default();
         map.extend(Self::load(Street::Turn).0);
         map.extend(Self::load(Street::Flop).0);
+        // TODO
+        // extend map with preflop
+        // alternatively, handle lossless preflop abstractions in Self::abstraction
         Self(map)
     }
 
@@ -43,6 +44,15 @@ impl Abstractor {
     }
     /// lookup the pre-computed abstraction for the outer observation
     pub fn abstraction(&self, outer: &Isomorphism) -> Abstraction {
+        // TODO
+        // match on street
+        // river => compute equity on the fly**
+        // turn | flop => lookup
+        // preflop => isomorphism into Hole
+        //
+        // ** this is expensive ?
+        // ** could implement mc_equity to not iterate over villain cards exhaustively ?
+        // ** should check benchmarks to see how much this matters
         self.0
             .get(outer)
             .cloned()

src/clustering/metric.rs

@@ -1,6 +1,5 @@
 use crate::clustering::abstraction::Abstraction;
 use crate::clustering::histogram::Histogram;
-// use crate::clustering::progress::Progress;
 use crate::clustering::xor::Pair;
 use std::collections::BTreeMap;
 

src/clustering/sampling.rs

@@ -3,11 +3,11 @@ use super::abstractor::Abstractor;
 use crate::cards::isomorphism::Isomorphism;
 use crate::mccfr::bucket::Bucket;
 use crate::mccfr::bucket::Path;
-use crate::mccfr::data::Vertex;
 use crate::mccfr::edge::Edge;
 use crate::mccfr::node::Node;
 use crate::mccfr::player::Player;
 use crate::mccfr::profile::Profile;
+use crate::mccfr::spot::Spot;
 use crate::play::game::Game;
 use crate::Probability;
 use rand::distributions::Distribution;
@@ -35,7 +35,7 @@ impl Sampler {
     /// compared to chance sampling, internal sampling, or full tree sampling.
     ///
     /// i think this could also be modified into a recursive CFR calcuation
-    pub fn sample(&self, node: &Node, profile: &Profile) -> Vec<(Vertex, Edge)> {
+    pub fn sample(&self, node: &Node, profile: &Profile) -> Vec<(Spot, Edge)> {
         let mut children = self.children(node);
         // terminal nodes have no children and we sample all possible actions for the traverser
         if node.player() == profile.walker() || children.is_empty() {
@@ -68,7 +68,7 @@ impl Sampler {
 
     /// produce the children of a Node.
     /// we may need some Trainer-level references to produce children
-    fn children(&self, node: &Node) -> Vec<(Vertex, Edge)> {
+    fn children(&self, node: &Node) -> Vec<(Spot, Edge)> {
         let ref game = node.datum().game();
         let ref past = node.history().into_iter().collect::<Vec<&Edge>>();
         game.children()
@@ -79,16 +79,16 @@ impl Sampler {
     }
     /// extend a path with an Edge
     /// wrap the (Game, Bucket) in a Data
-    fn explore(&self, game: Game, edge: Edge, history: &Vec<&Edge>) -> (Vertex, Edge) {
+    fn explore(&self, game: Game, edge: Edge, history: &Vec<&Edge>) -> (Spot, Edge) {
         let mut history = history.clone();
         history.push(&edge);
         (self.data(game, history), edge)
     }
     /// generate a Bucket from Game
     /// wrap the (Game, Bucket) in a Data
-    fn data(&self, game: Game, path: Vec<&Edge>) -> Vertex {
+    fn data(&self, game: Game, path: Vec<&Edge>) -> Spot {
         let bucket = self.bucket(&game, &path);
-        Vertex::from((game, bucket))
+        Spot::from((game, bucket))
     }
     /// inddd
     fn bucket(&self, game: &Game, path: &Vec<&Edge>) -> Bucket {

src/mccfr/mod.rs

@@ -1,10 +1,10 @@
 pub mod bucket;
-pub mod data;
 pub mod edge;
 pub mod info;
 pub mod memory;
 pub mod node;
 pub mod player;
 pub mod profile;
+pub mod spot;
 pub mod trainer;
 pub mod tree;

src/mccfr/node.rs

@@ -1,7 +1,7 @@
 use super::bucket::Bucket;
 use super::player::Player;
-use crate::mccfr::data::Vertex;
 use crate::mccfr::edge::Edge;
+use crate::mccfr::spot::Spot;
 use crate::play::continuation::Transition;
 use crate::Utility;
 use petgraph::graph::DiGraph;
@@ -13,11 +13,11 @@ use std::ptr::NonNull;
 pub struct Node {
     graph: NonNull<DiGraph<Self, Edge>>,
     index: NodeIndex,
-    datum: Vertex,
+    datum: Spot,
 }
 
-impl From<(NodeIndex, NonNull<DiGraph<Node, Edge>>, Vertex)> for Node {
-    fn from((index, graph, datum): (NodeIndex, NonNull<DiGraph<Node, Edge>>, Vertex)) -> Self {
+impl From<(NodeIndex, NonNull<DiGraph<Node, Edge>>, Spot)> for Node {
+    fn from((index, graph, datum): (NodeIndex, NonNull<DiGraph<Node, Edge>>, Spot)) -> Self {
         Self {
             index,
             graph,
@@ -28,7 +28,7 @@ impl From<(NodeIndex, NonNull<DiGraph<Node, Edge>>, Vertex)> for Node {
 
 /// collection of these three is what you would get in a Node, which may be too restrictive for a lot of the use so we'll se
 impl Node {
-    pub fn datum(&self) -> &Vertex {
+    pub fn datum(&self) -> &Spot {
         &self.datum
     }
     pub fn index(&self) -> NodeIndex {

src/mccfr/data.rs renamed to src/mccfr/spot.rs

@@ -4,18 +4,18 @@ use crate::mccfr::player::Player;
 use crate::play::continuation::Transition;
 use crate::play::game::Game;
 
-pub struct Vertex {
+pub struct Spot {
     game: Game,
     bucket: Bucket,
 }
 
-impl From<(Game, Bucket)> for Vertex {
+impl From<(Game, Bucket)> for Spot {
     fn from((game, bucket): (Game, Bucket)) -> Self {
         Self { game, bucket }
     }
 }
 
-impl Vertex {
+impl Spot {
     pub fn game(&self) -> &Game {
         &self.game
     }

src/mccfr/trainer.rs

@@ -1,8 +1,8 @@
-use super::data::Vertex;
 use super::edge::Edge;
 use super::node::Node;
 use super::player::Player;
 use super::profile::Profile;
+use super::spot::Spot;
 use super::tree::Tree;
 use crate::clustering::sampling::Sampler;
 use petgraph::graph::NodeIndex;
@@ -72,7 +72,7 @@ impl Explorer {
     /// recursively build the Tree from the given Node, according to the distribution defined by Profile.
     /// we assert the Tree property of every non-root Node having exactly one parent Edge
     /// we construct the appropriate references in self.attach() to ensure safety.
-    fn dfs(&mut self, head: Vertex, edge: Edge, root: NodeIndex) {
+    fn dfs(&mut self, head: Spot, edge: Edge, root: NodeIndex) {
         let head = self.witness(head);
         let head = self.tree.graph_mut().add_node(head);
         let edge = self.tree.graph_mut().add_edge(root, head, edge);
@@ -87,7 +87,7 @@ impl Explorer {
     /// attach a Node to the Tree,
     /// update the Profile to witness the new Node
     /// update the InfoPartition to witness the new Node.
-    fn witness(&mut self, data: Vertex) -> Node {
+    fn witness(&mut self, data: Spot) -> Node {
         let player = data.player().clone();
         let graph = self.tree.graph_ptr();
         let count = self.tree.graph_ref().node_count();
@@ -106,13 +106,13 @@ impl Explorer {
     /// Game::root() -> Observation -> Abstraction
     ///
     /// NOT deterministic, hole cards are thread_rng
-    fn root(&self) -> Vertex {
+    fn root(&self) -> Spot {
         use crate::mccfr::bucket::Bucket;
         use crate::play::game::Game;
         let node = Game::root();
         let path = self.sampler.path_abstraction(&Vec::new());
         let abstraction = self.sampler.card_abstraction(&node);
         let bucket = Bucket::from((path, abstraction));
-        Vertex::from((node, bucket))
+        Spot::from((node, bucket))
     }
 }