r2l-core 0.0.3

A rust reinforcement learning library
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

use candle_core::{Device, Tensor};
use itertools::izip;

use crate::{
    error::{Error, TensorError},
    tensor::{R2lTensor, VecTensor},
};

type Result<T> = std::result::Result<T, TensorError>;

impl From<candle_core::Error> for Error {
    fn from(error: candle_core::Error) -> Self {
        TensorError::operation("Candle backend", error).into()
    }
}

impl R2lTensor for Tensor {
    fn to_vec(&self) -> Result<Vec<f32>> {
        self.flatten_all()
            .and_then(|tensor| tensor.to_vec1())
            .map_err(|error| TensorError::operation("convert to vector", error))
    }

    fn to_shape(&self) -> Vec<usize> {
        self.shape().dims().to_vec()
    }

    fn from_slice_and_shape(data: &[f32], shape: Vec<usize>) -> Result<Self> {
        validate_shape(data.len(), &shape)?;
        Tensor::from_slice(data, shape, &Device::Cpu)
            .map_err(|error| TensorError::operation("construct from slice", error))
    }

    fn from_vec_and_shape(data: Vec<f32>, shape: Vec<usize>) -> Result<Self> {
        validate_shape(data.len(), &shape)?;
        Tensor::from_vec(data, shape, &Device::Cpu)
            .map_err(|error| TensorError::operation("construct from vector", error))
    }

    fn add(&self, other: &Self) -> Result<Self> {
        ensure_same_shape(self, other, "add")?;
        self.add(other)
            .map_err(|error| TensorError::operation("add", error))
    }

    fn sub(&self, other: &Self) -> Result<Self> {
        ensure_same_shape(self, other, "subtract")?;
        self.sub(other)
            .map_err(|error| TensorError::operation("subtract", error))
    }

    fn mul(&self, other: &Self) -> Result<Self> {
        ensure_same_shape(self, other, "multiply")?;
        self.mul(other)
            .map_err(|error| TensorError::operation("multiply", error))
    }

    fn exp(&self) -> Result<Self> {
        self.exp()
            .map_err(|error| TensorError::operation("exponential", error))
    }

    fn clamp(&self, min: f32, max: f32) -> Result<Self> {
        self.clamp(min, max)
            .map_err(|error| TensorError::operation("clamp", error))
    }

    fn minimum(&self, other: &Self) -> Result<Self> {
        ensure_same_shape(self, other, "minimum")?;
        Self::minimum(self, other).map_err(|error| TensorError::operation("minimum", error))
    }

    fn neg(&self) -> Result<Self> {
        self.neg()
            .map_err(|error| TensorError::operation("negate", error))
    }

    fn mean(&self) -> Result<Self> {
        if self.elem_count() == 0 {
            return Err(TensorError::EmptyInput {
                operation: "mean".into(),
            });
        }
        self.mean_all()
            .map_err(|error| TensorError::operation("mean", error))
    }

    fn sqr(&self) -> Result<Self> {
        self.sqr()
            .map_err(|error| TensorError::operation("square", error))
    }

    fn zeros(shape: Vec<usize>) -> Result<Self> {
        Tensor::zeros(shape, candle_core::DType::F32, &Device::Cpu)
            .map_err(|error| TensorError::operation("create zeros", error))
    }

    fn mul_scalar(&self, scalar: f32) -> Result<Self> {
        let scalar = Tensor::full(scalar, (), self.device())
            .map_err(|error| TensorError::operation("create scalar", error))?;
        self.broadcast_mul(&scalar)
            .map_err(|error| TensorError::operation("multiply by scalar", error))
    }
}

fn validate_shape(data_len: usize, shape: &[usize]) -> Result<()> {
    let expected = shape.iter().product();
    if expected != data_len {
        return Err(TensorError::InvalidShape {
            shape: shape.to_vec(),
            expected,
            actual: data_len,
        });
    }
    Ok(())
}

fn ensure_same_shape(left: &Tensor, right: &Tensor, operation: &str) -> Result<()> {
    let left = left.shape().dims().to_vec();
    let right = right.shape().dims().to_vec();
    if left != right {
        return Err(TensorError::ShapeMismatch {
            operation: operation.into(),
            left,
            right,
        });
    }
    Ok(())
}

impl VecTensor {
    /// Clamps each element between the corresponding values in `min` and `max`.
    ///
    /// # Errors
    ///
    /// Returns an error if the tensor shapes differ or the result cannot be constructed.
    pub fn clamp(&self, min: &Self, max: &Self) -> Result<Self> {
        self.ensure_same_shape(min, "clamp minimum")?;
        self.ensure_same_shape(max, "clamp maximum")?;
        let data = izip!(&self.data, &min.data, &max.data)
            .map(|(value, min, max)| value.clamp(*min, *max))
            .collect();
        Self::new(data, self.shape.clone())
    }
}