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
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166

use super::*;
use ndarray::{Array1, Ix1, Ix2, IntoDimension};
use std::ops::{Add, AddAssign, Mul, MulAssign};

type GradMap<C> = ::std::collections::HashMap<usize, C>;

#[derive(Clone, Debug, Default)]
#[cfg_attr(
    feature = "serde",
    derive(Serialize, Deserialize),
    serde(crate = "serde_crate")
)]
pub struct Columnar<C: Buffer<Dim = Ix1> = Array1<f64>> {
    dim: Ix2,
    grads: GradMap<C>,
}

impl<C: Buffer<Dim = Ix1>> Columnar<C> {
    pub fn new<D: IntoDimension<Dim = Ix2>>(dim: D, grads: GradMap<C>) -> Self {
        Columnar {
            dim: dim.into_dimension(),
            grads,
        }
    }

    pub fn from_column(n_cols: usize, index: usize, grad: C) -> Self {
        let dim = (grad.dim(), n_cols);
        let mut grads = GradMap::new();

        grads.insert(index, grad);

        Self::new(dim, grads)
    }
}

impl<C: Buffer<Dim = Ix1>> Buffer for Columnar<C> {
    type Dim = Ix2;

    fn dim(&self) -> (usize, usize) { self.dim.into_pattern() }

    fn raw_dim(&self) -> Ix2 { self.dim }

    fn addto<D: DataMut<Elem = f64>>(&self, weights: &mut ArrayBase<D, Ix2>) {
        for (&c, pds) in self.grads.iter() {
            pds.addto(&mut weights.column_mut(c));
        }
    }

    fn scaled_addto<D: DataMut<Elem = f64>>(&self, alpha: f64, weights: &mut ArrayBase<D, Ix2>) {
        for (&c, buf) in self.grads.iter() {
            buf.scaled_addto(alpha, &mut weights.column_mut(c));
        }
    }

    fn to_dense(&self) -> Array2<f64> {
        let mut g_matrix = Array2::zeros(self.dim);

        self.addto(&mut g_matrix);

        g_matrix
    }
}

impl<C: BufferMut<Dim = Ix1>> BufferMut for Columnar<C> {
    fn zeros<D: IntoDimension<Dim = Ix2>>(dim: D) -> Self { Self::new(dim, GradMap::new()) }

    fn map(&self, f: impl Fn(f64) -> f64) -> Self {
        Columnar {
            dim: self.dim,
            grads: self.grads.iter().map(|(c, j)| (*c, j.map(&f))).collect(),
        }
    }

    fn map_into(mut self, f: impl Fn(f64) -> f64) -> Self {
        self.map_inplace(f);
        self
    }

    fn map_inplace(&mut self, f: impl Fn(f64) -> f64) {
        self.grads.values_mut().for_each(|buf| buf.map_inplace(&f));
    }

    fn merge(&self, other: &Self, f: impl Fn(f64, f64) -> f64) -> Self {
        Columnar {
            dim: self.dim,
            grads: self
                .grads
                .iter()
                .map(|(c, j1)| match other.grads.get(c) {
                    Some(j2) => (*c, j1.merge(j2, &f)),
                    None => (*c, j1.map(|x| f(x, 0.0))),
                })
                .collect(),
        }
    }

    fn merge_inplace(&mut self, other: &Self, f: impl Fn(f64, f64) -> f64) {
        for (c, xs) in self.grads.iter_mut() {
            if let Some(ys) = other.grads.get(c) {
                xs.merge_inplace(ys, &f)
            } else {
                xs.map_inplace(|x| f(x, 0.0))
            }
        }

        for (&c, ys) in other.grads.iter() {
            self.grads.entry(c).or_insert_with(|| ys.map(|y| f(0.0, y)));
        }
    }
}

impl<C: Buffer<Dim = Ix1>> Into<Array2<f64>> for Columnar<C> {
    fn into(self) -> Array2<f64> { self.to_dense() }
}

impl<C: BufferMut<Dim = Ix1>> Add<Columnar<C>> for Columnar<C> {
    type Output = Columnar<C>;

    fn add(mut self, other: Columnar<C>) -> Columnar<C> {
        self.add_assign(&other);

        self
    }
}

impl<C: BufferMut<Dim = Ix1>> AddAssign<Columnar<C>> for Columnar<C> {
    fn add_assign(&mut self, other: Columnar<C>) {
        for (c, buf_other) in other.grads.into_iter() {
            if let Some(buf) = self.grads.get_mut(&c) {
                buf.merge_inplace(&buf_other, |x, y| x + y);
            } else {
                self.grads.insert(c, buf_other);
            }
        }
    }
}

impl<C: BufferMut<Dim = Ix1>> AddAssign<&Columnar<C>> for Columnar<C> {
    fn add_assign(&mut self, other: &Columnar<C>) {
        for (&c, buf_other) in other.grads.iter() {
            if let Some(buf) = self.grads.get_mut(&c) {
                buf.merge_inplace(buf_other, |x, y| x + y);
            } else {
                self.grads.insert(c, buf_other.to_owned());
            }
        }
    }
}

impl<C: BufferMut<Dim = Ix1>> Mul<f64> for Columnar<C> {
    type Output = Columnar<C>;

    fn mul(mut self, alpha: f64) -> Columnar<C> {
        self.mul_assign(alpha);

        self
    }
}

impl<C: BufferMut<Dim = Ix1>> MulAssign<f64> for Columnar<C> {
    fn mul_assign(&mut self, alpha: f64) {
        for buf in self.grads.values_mut() {
            buf.map_inplace(|x| alpha * x);
        }
    }
}