def __init__(self, width, ker_width, depth, ker_in, in_width=1, out_width=1):
super(KernelNNBoundary, self).__init__()
self.depth = depth
self.fc1 = torch.nn.Linear(in_width, width)
kernel = DenseNet([ker_in, ker_width//2, ker_width, width**2], torch.nn.ReLU)
self.conv1 = NNConv_old(width, width, kernel, aggr='mean')
kernel2 = DenseNet([ker_in, ker_width//2, ker_width, width**2], torch.nn.ReLU)
self.conv2 = NNConv_old(width, width, kernel2, aggr='mean')
self.fc2 = torch.nn.Linear(width, ker_width)
self.fc3 = torch.nn.Linear(ker_width, 1)
def __init__(self):
super(Net_MP, self).__init__()
self.fc1 = torch.nn.Linear(3, width)
kernel = nn.Sequential(nn.Linear(3, width), nn.ReLU(), nn.Linear(width, width**2))
self.conv1 = NNConv_old(width, width, kernel, aggr='mean')
self.fc2 = torch.nn.Linear(width, 1)
def __init__(self, width_node, width_kernel, depth, ker_in, in_width=1, out_width=1):
super(KernelNN3, self).__init__()
self.depth = depth
self.fc1 = torch.nn.Linear(in_width, width_node)
kernel = DenseNet([ker_in, width_kernel // 2, width_kernel, width_node**2], torch.nn.ReLU)
self.conv1 = NNConv_old(width_node, width_node, kernel, aggr='mean')
self.fc2 = torch.nn.Linear(width_node, 1)
def __init__(self,
width,
width_mid,
ker_width,
depth,
ker_in,
in_width=1,
out_width=1):
super(KernelNN, self).__init__()
self.depth = depth
self.fc1 = torch.nn.Linear(in_width, width)
kernel1 = DenseNet([ker_in, ker_width, ker_width, width**2],
torch.nn.ReLU)
self.conv1 = NNConv_old(width, width, kernel1, aggr='mean')
kernel2 = DenseNet([ker_in, ker_width, ker_width, width_mid**2],
torch.nn.ReLU)
self.conv2 = NNConv_old(width_mid, width_mid, kernel2, aggr='mean')
kernel3 = DenseNet([ker_in, ker_width, ker_width, width**2],
torch.nn.ReLU)
self.conv3 = NNConv_old(width, width, kernel3, aggr='mean')
kernel12 = DenseNet([ker_in, ker_width, ker_width, width * width_mid],
torch.nn.ReLU)
self.conv12 = NNConv_old(width, width_mid, kernel12, aggr='mean')
kernel23 = DenseNet([ker_in, ker_width, ker_width, width * width_mid],
torch.nn.ReLU)
self.conv23 = NNConv_old(width_mid, width, kernel23, aggr='mean')
kernel32 = DenseNet([ker_in, ker_width, ker_width, width * width_mid],
torch.nn.ReLU)
self.conv32 = NNConv_old(width, width_mid, kernel32, aggr='mean')
kernel21 = DenseNet([ker_in, ker_width, ker_width, width * width_mid],
torch.nn.ReLU)
self.conv21 = NNConv_old(width_mid, width, kernel21, aggr='mean')
self.fc2 = torch.nn.Linear(width, 1)
def __init__(self):
super(Net_MP_one, self).__init__()
kernel = nn.Sequential(nn.Linear(3, width), nn.ReLU(), nn.Linear(width, 1))
self.conv1 = NNConv_old(1, 1, kernel, aggr='mean')