File size: 5,274 Bytes
37e9e4f |
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 |
import torch
import torch.nn as nn
import torch.nn.functional as F
import math
class KANLinear(nn.Module):
def __init__(self, in_features, out_features, grid_size=5, spline_order=3, scale_noise=0.1, scale_base=1.0, scale_spline=1.0, enable_standalone_scale_spline=True, base_activation=nn.SiLU, grid_eps=0.02, grid_range=[-1, 1]):
super(KANLinear, self).__init__()
self.in_features = in_features
self.out_features = out_features
self.grid_size = grid_size
self.spline_order = spline_order
h = (grid_range[1] - grid_range[0]) / grid_size
grid = ((torch.arange(-spline_order, grid_size + spline_order + 1) * h + grid_range[0]).expand(in_features, -1).contiguous())
self.register_buffer("grid", grid)
self.base_weight = nn.Parameter(torch.Tensor(out_features, in_features))
self.spline_weight = nn.Parameter(torch.Tensor(out_features, in_features, grid_size + spline_order))
if enable_standalone_scale_spline:
self.spline_scaler = nn.Parameter(torch.Tensor(out_features, in_features))
self.scale_noise = scale_noise
self.scale_base = scale_base
self.scale_spline = scale_spline
self.enable_standalone_scale_spline = enable_standalone_scale_spline
self.base_activation = base_activation()
self.grid_eps = grid_eps
self.reset_parameters()
def reset_parameters(self):
nn.init.kaiming_uniform_(self.base_weight, a=math.sqrt(5) * self.scale_base)
with torch.no_grad():
noise = ((torch.rand(self.grid_size + 1, self.in_features, self.out_features) - 1 / 2) * self.scale_noise / self.grid_size)
self.spline_weight.data.copy_((self.scale_spline if not self.enable_standalone_scale_spline else 1.0) * self.curve2coeff(self.grid.T[self.spline_order : -self.spline_order], noise))
if self.enable_standalone_scale_spline:
nn.init.kaiming_uniform_(self.spline_scaler, a=math.sqrt(5) * self.scale_spline)
def b_splines(self, x: torch.Tensor):
assert x.dim() == 2 and x.size(1) == self.in_features
grid = self.grid
x = x.unsqueeze(-1)
bases = ((x >= grid[:, :-1]) & (x < grid[:, 1:])).to(x.dtype)
for k in range(1, self.spline_order + 1):
bases = ((x - grid[:, : -(k + 1)]) / (grid[:, k:-1] - grid[:, : -(k + 1)]) * bases[:, :, :-1]) + ((grid[:, k + 1 :] - x) / (grid[:, k + 1 :] - grid[:, 1:(-k)]) * bases[:, :, 1:])
assert bases.size() == (x.size(0), self.in_features, self.grid_size + self.spline_order)
return bases.contiguous()
def curve2coeff(self, x: torch.Tensor, y: torch.Tensor):
assert x.dim() == 2 and x.size(1) == self.in_features
assert y.size() == (x.size(0), self.in_features, self.out_features)
A = self.b_splines(x).transpose(0, 1)
B = y.transpose(0, 1)
solution = torch.linalg.lstsq(A, B).solution
result = solution.permute(2, 0, 1)
assert result.size() == (self.out_features, self.in_features, self.grid_size + self.spline_order)
return result.contiguous()
@property
def scaled_spline_weight(self):
return self.spline_weight * (self.spline_scaler.unsqueeze(-1) if self.enable_standalone_scale_spline else 1.0)
def forward(self, x: torch.Tensor):
assert x.dim() == 2 and x.size(1) == self.in_features
base_output = F.linear(self.base_activation(x), self.base_weight)
spline_output = F.linear(self.b_splines(x).view(x.size(0), -1), self.scaled_spline_weight.view(self.out_features, -1))
return base_output + spline_output
@torch.no_grad()
def update_grid(self, x: torch.Tensor, margin=0.01):
assert x.dim() == 2 and x.size(1) == self.in_features
batch = x.size(0)
splines = self.b_splines(x).permute(1, 0, 2)
orig_coeff = self.scaled_spline_weight.permute(1, 2, 0)
unreduced_spline_output = torch.bmm(splines, orig_coeff).permute(1, 0, 2)
x_sorted = torch.sort(x, dim=0)[0]
grid_adaptive = x_sorted[torch.linspace(0, batch - 1, self.grid_size + 1, dtype=torch.int64, device=x.device)]
uniform_step = (x_sorted[-1] - x_sorted[0] + 2 * margin) / self.grid_size
grid_uniform = (torch.arange(self.grid_size + 1, dtype=torch.float32, device=x.device).unsqueeze(1) * uniform_step + x_sorted[0] - margin)
grid = self.grid_eps * grid_uniform + (1 - self.grid_eps) * grid_adaptive
grid = torch.cat([grid[:1] - uniform_step * torch.arange(self.spline_order, 0, -1, device=x.device).unsqueeze(1), grid, grid[-1:] + uniform_step * torch.arange(1, self.spline_order + 1, device=x.device).unsqueeze(1)], dim=0)
self.grid.copy_(grid.T)
self.spline_weight.data.copy_(self.curve2coeff(x, unreduced_spline_output))
def regularization_loss(self, regularize_activation=1.0, regularize_entropy=1.0):
l1_fake = self.spline_weight.abs().mean(-1)
regularization_loss_activation = l1_fake.sum()
p = l1_fake / regularization_loss_activation
regularization_loss_entropy = -torch.sum(p * p.log())
return regularize_activation * regularization_loss_activation + regularize_entropy * regularization_loss_entropy
|