Skip to content

Sine

continuiti.benchmarks.sine

Sine benchmarks.

SineBenchmark(n_sensors=32, n_evaluations=32, n_train=1024, n_test=32, uniform=False)

Bases: Benchmark

Sine benchmark.

The SineBenchmark contains a dataset of trigonometric functions

\[ f_k(x) = \sin(kx), \quad x \in [-1, 1], \quad k \in [\pi, 2\pi], \]

with the following properties:

  • Input and output function spaces are the same.
  • The input space is mapped to the output space with the identity operator.
  • Both the the domain and the co-domain are sampled on a regular grid, if uniform is False, or uniformly, otherwise.
  • The parameter \(k\) is sampled uniformly from \([\pi, 2\pi]\).
PARAMETER DESCRIPTION
n_sensors

number of sensors.

TYPE: int DEFAULT: 32

n_evaluations

number of evaluations.

TYPE: int DEFAULT: 32

n_train

number of observations in the train dataset.

TYPE: int DEFAULT: 1024

n_test

number of observations in the test dataset.

TYPE: int DEFAULT: 32

uniform

whether to sample the domain and co-domain random uniformly.

TYPE: bool DEFAULT: False

Source code in src/continuiti/benchmarks/sine.py 
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
def __init__(
    self,
    n_sensors: int = 32,
    n_evaluations: int = 32,
    n_train: int = 1024,
    n_test: int = 32,
    uniform: bool = False,
):
    sine_set = FunctionSet(lambda k: Function(lambda x: torch.sin(k * x)))

    if uniform:
        x_sampler = y_sampler = UniformBoxSampler([-1.0], [1.0])
    else:
        x_sampler = y_sampler = RegularGridSampler([-1.0], [1.0])

    parameter_sampler = UniformBoxSampler([torch.pi], [2 * torch.pi])

    def get_dataset(n_obs: int):
        return FunctionOperatorDataset(
            input_function_set=sine_set,
            x_sampler=x_sampler,
            n_sensors=n_sensors,
            output_function_set=sine_set,
            y_sampler=y_sampler,
            n_evaluations=n_evaluations,
            parameter_sampler=parameter_sampler,
            n_observations=n_obs,
        )

    train_dataset = get_dataset(n_train)
    test_dataset = get_dataset(n_test)

    super().__init__(train_dataset, test_dataset, [MSELoss()])

SineRegular()

Bases: SineBenchmark

Sine benchmark with the domain and co-domain sampled on a regular grid.

The SineRegular benchmark is a SineBenchmark with the following properties:

  • n_sensors is 32.
  • n_evaluations is 32.
  • n_train is 1024.
  • n_test is 1024.
  • uniform is False.

Visualizations of a few samples.

Source code in src/continuiti/benchmarks/sine.py 
90
91
92
93
94
95
96
97
def __init__(self):
    super().__init__(
        n_sensors=32,
        n_evaluations=32,
        n_train=1024,
        n_test=1024,
        uniform=False,
    )

SineUniform()

Bases: SineBenchmark

Sine benchmark with the domain and co-domain sampled random uniformly.

The SineRegular benchmark is a SineBenchmark with the following properties:

  • n_sensors is 32.
  • n_evaluations is 32.
  • n_train is 4096.
  • n_test is 4096.
  • uniform is True.

Visualizations of a few samples.

Source code in src/continuiti/benchmarks/sine.py 
116
117
118
119
120
121
122
123
def __init__(self):
    super().__init__(
        n_sensors=32,
        n_evaluations=32,
        n_train=4096,
        n_test=4096,
        uniform=True,
    )
Last update: 2024-08-22
Created: 2024-08-22