block_graph/builtin/
reinforcement.rs

1impl AI<Vec<f32>,Vec<f32>> for AccQLayer{
2	fn forward(&self,mut input:Vec<f32>)->Vec<f32>{
3		let (dim,gamma)=(self.dim,self.gamma);
4		assert!(dim==0||dim==-1,"Dimension index was {dim} but a vec only has one tensor dimension");
5
6		input.iter_mut().rev().fold(0.0,|future,present|{
7			*present+=future*gamma;
8			*present
9		});
10		input
11	}
12}
13impl AccQLayer{
14	/// creates from the inner value
15	pub fn new(gamma:f32)->Self{Self::default().with_gamma(gamma)}
16	/// gets the dimension
17	pub fn get_dim(&self)->i32{self.dim}
18	/// gets the gamma
19	pub fn get_gamma(&self)->f32{self.gamma}
20	/// sets the dim
21	pub fn set_dim(&mut self,dim:i32){self.dim=dim}
22	/// sets the gamma
23	pub fn set_gamma(&mut self,gamma:f32){self.gamma=gamma}
24	/// sets the dim
25	pub fn with_dim(mut self,dim:i32)->Self{
26		self.dim=dim;
27		self
28	}
29	/// withs the gamma
30	pub fn with_gamma(mut self,gamma:f32)->Self{
31		self.gamma=gamma;
32		self
33	}
34}
35impl Decompose for AccQLayer{
36	fn compose((dim,gamma):Self::Decomposition)->Self{
37		Self{dim,gamma}
38	}
39	fn decompose(self)->Self::Decomposition{(self.dim,self.gamma)}
40	fn decompose_cloned(&self)->Self::Decomposition{(self.dim,self.gamma)}
41	type Decomposition=(i32,f32);
42}
43impl Default for AccQLayer{
44	fn default()->Self{
45		Self{dim:-1,gamma:0.9}
46	}
47}
48impl Op for AccQLayer{
49	type Output=Vec<f32>;
50}
51impl<A:AI<X,Y>,X,Y> AI<X,Y> for AccQ<A> where AccQLayer:AI<Y,Y>{
52	fn forward(&self,input:X)->Y{self.layer.forward(self.inner.forward(input))}
53	fn forward_mut(&mut self,input:X)->Y{self.layer.forward(self.inner.forward_mut(input))}
54}
55impl<A:Decompose> Decompose for AccQ<A>{
56	fn compose((inner,(dim,gamma)):Self::Decomposition)->Self{
57		Self{inner:A::compose(inner),layer:AccQLayer::compose((dim,gamma))}
58	}
59	fn decompose(self)->Self::Decomposition{(self.inner.decompose(),self.layer.decompose())}
60	fn decompose_cloned(&self)->Self::Decomposition{(self.inner.decompose_cloned(),self.layer.decompose_cloned())}
61	type Decomposition=(A::Decomposition,(i32,f32));
62}
63impl<A:IntoSequence<M>,M:AI<M::Output,M::Output>+Op> IntoSequence<M> for AccQ<A> where AccQLayer:Into<M>{
64	fn into_sequence(self)->Sequential<Vec<M>>{self.inner.into_sequence().with_next(self.layer)}
65}
66impl<A:Op<Output=Y>,Y> Op for AccQ<A> where AccQLayer:AI<Y,Y>{
67	type Output=Y;
68}
69impl<A:UnwrapInner> UnwrapInner for AccQ<A>{
70	fn unwrap_inner(self)->Self::Inner{self.into_inner().unwrap_inner()}
71	type Inner=A::Inner;
72}
73impl<A> AccQ<A>{
74	/// gets the dimension
75	pub fn get_dim(&self)->i32{self.layer.dim}
76	/// gets the gamma
77	pub fn get_gamma(&self)->f32{self.layer.gamma}
78	/// references the inner value
79	pub fn inner(&self)->&A{&self.inner}
80	/// references the inner value
81	pub fn inner_mut(&mut self)->&mut A{&mut self.inner}
82	/// converts into the inner value
83	pub fn into_inner(self)->A{self.inner}
84	/// creates from the inner value
85	pub fn new(gamma:f32,inner:A)->Self{
86		let layer=AccQLayer::new(gamma);
87		Self{inner,layer}
88	}
89	/// sets the dim
90	pub fn set_dim(&mut self,dim:i32){self.layer.dim=dim}
91	/// sets the gamma
92	pub fn set_gamma(&mut self,gamma:f32){self.layer.gamma=gamma}
93	/// sets the dim
94	pub fn with_dim(mut self,dim:i32)->Self{
95		self.layer.dim=dim;
96		self
97	}
98	/// withs the gamma
99	pub fn with_gamma(mut self,gamma:f32)->Self{
100		self.layer.gamma=gamma;
101		self
102	}
103	/// replaces the inner value
104	pub fn with_inner<B>(&self,inner:B)->AccQ<B> where AccQ<B>:Op{AccQ::new(self.get_gamma(),inner).with_dim(self.get_dim())}
105}
106impl<M:AI<M::Output,M::Output>+Op> IntoSequence<M> for AccQLayer where AccQLayer:Into<M>{
107	fn into_sequence(self)->Sequential<Vec<M>>{vec![self.into()].sequential()}
108}
109#[derive(Clone,Copy,Debug,Deserialize,Default,PartialEq,Serialize)]
110/// accumulates cumulative
111pub struct AccQ<A>{inner:A,layer:AccQLayer}
112#[derive(Clone,Copy,Debug,Deserialize,PartialEq,Serialize)]
113/// accumulates cumulative
114pub struct AccQLayer{dim:i32,gamma:f32}
115use crate::{
116	AI,Decompose,IntoSequence,Op,UnwrapInner
117};
118use serde::{Deserialize,Serialize};
119use super::Sequential;
block_graph/builtin/reinforcement.rs

block_graph/builtin/
reinforcement.rs
