rsrl 0.8.1

A fast, extensible reinforcement learning framework in Rust
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142

use crate::{
    domains::Transition,
    fa::StateActionUpdate,
    policies::Policy,
    Enumerable,
    Function,
    Handler,
};
use std::f64;

// TODO: Extract prediction component GQ / GQ(lambda) into seperate
// implementations.

#[derive(Clone, Copy, Debug)]
#[cfg_attr(
    feature = "serde",
    derive(Serialize, Deserialize),
    serde(crate = "serde_crate")
)]
pub struct Response<RQ, RT> {
    pub td_error: f64,

    pub q_response: RQ,
    pub td_response: RT,
}

#[derive(Clone, Copy, Debug)]
#[cfg_attr(
    feature = "serde",
    derive(Serialize, Deserialize),
    serde(crate = "serde_crate")
)]
pub enum Error<EQ, ET> {
    QFuncError(EQ),
    TDEstError(ET),
}

/// Greedy GQ control algorithm.
///
/// Maei, Hamid R., et al. "Toward off-policy learning control with function
/// approximation." Proceedings of the 27th International Conference on Machine
/// Learning (ICML-10). 2010.
#[derive(Clone, Debug, Parameterised)]
#[cfg_attr(
    feature = "serde",
    derive(Serialize, Deserialize),
    serde(crate = "serde_crate")
)]
pub struct GreedyGQ<Q, T, P> {
    #[weights]
    pub fa_q: Q,
    pub fa_td: T,

    pub behaviour_policy: P,

    pub gamma: f64,
}

impl<'m, S, Q, T, P> Handler<&'m Transition<S, P::Action>> for GreedyGQ<Q, T, P>
where
    Q: Handler<StateActionUpdate<&'m S, usize>> + Function<(&'m S,)> + Enumerable<(&'m S,)>,

    Q::Output: IntoIterator<Item = f64> + std::ops::Index<usize, Output = f64>,
    <Q::Output as IntoIterator>::IntoIter: ExactSizeIterator,

    T: Handler<StateActionUpdate<&'m S, usize>> + Function<(&'m S, usize), Output = f64>,

    P: Policy<&'m S, Action = usize>,
{
    type Response = Response<Q::Response, T::Response>;
    type Error = Error<Q::Error, T::Error>;

    fn handle(&mut self, t: &'m Transition<S, P::Action>) -> Result<Self::Response, Self::Error> {
        let s = t.from.state();

        let qsa = self.fa_q.evaluate_index((s,), t.action);
        let td_est = self.fa_td.evaluate((s, t.action));

        if t.terminated() {
            let td_error = t.reward - qsa;

            let q_response = self.fa_q
                .handle(StateActionUpdate {
                    state: s,
                    action: t.action,
                    error: td_error,
                })
                .map_err(|e| Error::QFuncError(e))?;

            let td_response = self.fa_td
                .handle(StateActionUpdate {
                    state: s,
                    action: t.action,
                    error: td_error - td_est,
                })
                .map_err(|e| Error::TDEstError(e))?;

            Ok(Response {
                td_error,

                q_response,
                td_response,
            })
        } else {
            let ns = t.to.state();
            let (na, qnsna) = self.fa_q.find_max((ns,));

            let td_error = t.reward + self.gamma * qnsna - qsa;

            let q_response = self.fa_q
                .handle(StateActionUpdate {
                    state: s,
                    action: t.action,
                    error: td_error,
                })
                .and_then(|_| {
                    self.fa_q
                        .handle(StateActionUpdate {
                            state: ns,
                            action: na,
                            error: -self.gamma * td_est,
                        })
                })
                .map_err(|e| Error::QFuncError(e))?;

            let td_response = self.fa_td
                .handle(StateActionUpdate {
                    state: s,
                    action: t.action,
                    error: td_error - td_est,
                })
                .map_err(|e| Error::TDEstError(e))?;

            Ok(Response {
                td_error,

                q_response,
                td_response,
            })
        }
    }
}