File size: 7,388 Bytes
4a96652
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
from __future__ import annotations

from collections.abc import Mapping, Sequence
from pathlib import Path

import numpy as np
import torch
import torch.nn as nn
from scipy.special import gamma, jacobi, roots_jacobi


def jacobi_poly(
    degree: int, alpha: float, beta: float, points: np.ndarray | float
) -> np.ndarray:
    return jacobi(degree, alpha, beta)(np.asarray(points, dtype=float))


def gauss_lobatto_jacobi_weights(
    order: int, alpha: float = 0.0, beta: float = 0.0
) -> tuple[np.ndarray, np.ndarray]:
    """Return Gauss-Lobatto-Jacobi nodes and weights, including both endpoints."""
    if order < 3:
        raise ValueError("Gauss-Lobatto-Jacobi quadrature order must be at least 3")

    interior = roots_jacobi(order - 2, alpha + 1, beta + 1)[0]
    polynomial = jacobi_poly(order - 1, alpha, beta, interior)
    if alpha == 0 and beta == 0:
        weights = 2.0 / ((order - 1) * order * np.square(polynomial))
        left_weight = 2.0 / (
            (order - 1)
            * order
            * np.square(jacobi_poly(order - 1, 0, 0, -1.0))
        )
        right_weight = 2.0 / (
            (order - 1)
            * order
            * np.square(jacobi_poly(order - 1, 0, 0, 1.0))
        )
    else:
        denominator = (
            (order - 1)
            * gamma(order)
            * gamma(alpha + beta + order + 1)
            * np.square(polynomial)
        )
        numerator = (
            2.0 ** (alpha + beta + 1)
            * gamma(alpha + order)
            * gamma(beta + order)
        )
        weights = numerator / denominator
        left_weight = (
            (beta + 1)
            * numerator
            / (
                (order - 1)
                * gamma(order)
                * gamma(alpha + beta + order + 1)
                * np.square(jacobi_poly(order - 1, alpha, beta, -1.0))
            )
        )
        right_weight = (
            (alpha + 1)
            * numerator
            / (
                (order - 1)
                * gamma(order)
                * gamma(alpha + beta + order + 1)
                * np.square(jacobi_poly(order - 1, alpha, beta, 1.0))
            )
        )

    nodes = np.concatenate(([-1.0], interior, [1.0]))
    all_weights = np.concatenate(([left_weight], weights, [right_weight]))
    return nodes, all_weights


def test_function_jacobi(
    degree: int, points: np.ndarray, alpha: float = 0.0, beta: float = 0.0
) -> np.ndarray:
    """Evaluate phi_n = P_(n+1) - P_(n-1), which vanishes at both endpoints."""
    if degree < 1:
        raise ValueError("test function degree must be at least 1")
    return jacobi_poly(degree + 1, alpha, beta, points) - jacobi_poly(
        degree - 1, alpha, beta, points
    )


def d_test_function_jacobi(
    degree: int, points: np.ndarray
) -> tuple[np.ndarray, np.ndarray]:
    """Evaluate the first and second derivatives of a Jacobi test function."""
    if degree < 1:
        raise ValueError("test function degree must be at least 1")
    points = np.asarray(points, dtype=float)
    if degree == 1:
        first = ((degree + 2) / 2) * jacobi_poly(degree, 1, 1, points)
        second = ((degree + 2) * (degree + 3) / 4) * jacobi_poly(
            degree - 1, 2, 2, points
        )
    elif degree == 2:
        first = ((degree + 2) / 2) * jacobi_poly(
            degree, 1, 1, points
        ) - (degree / 2) * jacobi_poly(degree - 2, 1, 1, points)
        second = ((degree + 2) * (degree + 3) / 4) * jacobi_poly(
            degree - 1, 2, 2, points
        )
    else:
        first = ((degree + 2) / 2) * jacobi_poly(
            degree, 1, 1, points
        ) - (degree / 2) * jacobi_poly(degree - 2, 1, 1, points)
        second = ((degree + 2) * (degree + 3) / 4) * jacobi_poly(
            degree - 1, 2, 2, points
        ) - (degree * (degree + 1) / 4) * jacobi_poly(
            degree - 3, 2, 2, points
        )
    return np.asarray(first), np.asarray(second)


class VPINN(nn.Module):
    """Fully connected VPINN backbone with trainable adaptive activations."""

    def __init__(
        self,
        layers: Sequence[int],
        activation: str = "tanh",
        dtype: torch.dtype = torch.float64,
    ) -> None:
        super().__init__()
        if len(layers) < 2:
            raise ValueError("layers must contain at least an input and an output size")
        if activation not in {"tanh", "sin", "cos"}:
            raise ValueError(f"unsupported activation: {activation}")

        self.activation = activation
        self.linears = nn.ModuleList(
            nn.Linear(layers[index], layers[index + 1], dtype=dtype)
            for index in range(len(layers) - 1)
        )
        # Keep one value per layer for compatibility with the bundled checkpoint.
        self.a = nn.ParameterList(
            nn.Parameter(torch.tensor(0.05, dtype=dtype))
            for _ in range(len(layers) - 1)
        )
        for linear in self.linears:
            nn.init.xavier_normal_(linear.weight)
            nn.init.zeros_(linear.bias)

    def forward(self, inputs: torch.Tensor) -> torch.Tensor:
        hidden = inputs
        activation = getattr(torch, self.activation)
        for index, linear in enumerate(self.linears[:-1]):
            hidden = activation(20.0 * self.a[index] * linear(hidden))
        return self.linears[-1](hidden)


def unpack_checkpoint(checkpoint: Mapping) -> tuple[Mapping[str, torch.Tensor], dict]:
    """Support both legacy raw state dictionaries and metadata checkpoints."""
    if "model_state" in checkpoint:
        return checkpoint["model_state"], dict(checkpoint)
    if checkpoint and all(torch.is_tensor(value) for value in checkpoint.values()):
        return checkpoint, {}
    raise ValueError("checkpoint does not contain a valid VPINN state dictionary")


def infer_layers(state: Mapping[str, torch.Tensor]) -> list[int]:
    weight_keys = sorted(
        (key for key in state if key.startswith("linears.") and key.endswith(".weight")),
        key=lambda key: int(key.split(".")[1]),
    )
    if not weight_keys:
        raise ValueError("checkpoint contains no VPINN linear-layer weights")
    layers = [int(state[weight_keys[0]].shape[1])]
    layers.extend(int(state[key].shape[0]) for key in weight_keys)
    return layers


class VPINNWrapper(nn.Module):
    """Load a VPINN checkpoint and expose it as an inference module."""

    def __init__(
        self,
        weight_path: str | Path,
        device: str | torch.device | None = None,
        dtype: torch.dtype = torch.float64,
    ) -> None:
        super().__init__()
        resolved_device = torch.device(
            device or ("cuda" if torch.cuda.is_available() else "cpu")
        )
        checkpoint = torch.load(weight_path, map_location="cpu", weights_only=True)
        if not isinstance(checkpoint, Mapping):
            raise ValueError(f"invalid VPINN checkpoint: {weight_path}")
        state, metadata = unpack_checkpoint(checkpoint)
        layers = metadata.get("layers") or infer_layers(state)
        self.device = resolved_device
        self.dtype = dtype
        self.net = VPINN(layers, dtype=dtype).to(device=resolved_device, dtype=dtype)
        self.net.load_state_dict(state, strict=True)
        self.net.eval()

    def forward(self, inputs: torch.Tensor) -> torch.Tensor:
        return self.net(inputs.to(device=self.device, dtype=self.dtype))