def __init__(self,
class_count,
n_features=3,
num_points=1024,
sort_pool_k=32):
super(PointNet2MRGSortPoolClass, self).__init__()
nFeaturesL2 = 3 + 128
shared_mpls = [
SAModuleFullPoint(0.4, 16, MLP([n_features, 64, 64, 128])),
SAModuleFullPoint(0.9, 32, MLP([nFeaturesL2, 128, 128, 256]))
]
# The mpls are shared to lower the model memory footprint
self.high_resolution_module = SAModuleMRG(num_points, 512, shared_mpls)
self.mid_resolution_module = SAModuleMRG(num_points, 256, shared_mpls)
self.low_resolution_module = SAModuleMRG(num_points, 128, shared_mpls)
self.readout = GlobalSortPool(MLP([789, 1024, 1024, 1024]),
k=sort_pool_k)
# Classification Layers
sort_pool_out = 1024 * sort_pool_k
self.lin1 = Lin(sort_pool_out, 512)
self.bn1 = nn.BatchNorm1d(512)
self.lin2 = Lin(512, 256)
self.bn2 = nn.BatchNorm1d(256)
self.lin3 = Lin(256, class_count)
def test_dense_sage_conv():
in_channels, out_channels = (16, 32)
nn = Seq(Lin(in_channels, 32), ReLU(), Lin(32, out_channels))
sparse_conv = GINConv(nn)
dense_conv = DenseGINConv(nn)
dense_conv = DenseGINConv(nn, train_eps=True)
assert dense_conv.__repr__() == (
'DenseGINConv(nn=Sequential(\n'
' (0): Linear(in_features=16, out_features=32, bias=True)\n'
' (1): ReLU()\n'
' (2): Linear(in_features=32, out_features=32, bias=True)\n'
'))')
x = torch.randn((5, in_channels))
edge_index = torch.tensor([[0, 0, 1, 1, 2, 2, 3, 4],
[1, 2, 0, 2, 0, 1, 4, 3]])
sparse_out = sparse_conv(x, edge_index)
x = torch.cat([x, x.new_zeros(1, in_channels)],
dim=0).view(2, 3, in_channels)
adj = torch.Tensor([
[[0, 1, 1], [1, 0, 1], [1, 1, 0]],
[[0, 1, 0], [1, 0, 0], [0, 0, 0]],
])
mask = torch.tensor([[1, 1, 1], [1, 1, 0]], dtype=torch.uint8)
dense_out = dense_conv(x, adj, mask)
assert dense_out.size() == (2, 3, out_channels)
dense_out = dense_out.view(6, out_channels)[:-1]
assert torch.allclose(sparse_out, dense_out, atol=1e-04)
def __init__(self, in_channels, out_channels, dim_model, k=16):
super().__init__()
self.k = k
# dummy feature is created if there is none given
in_channels = max(in_channels, 1)
# first block
self.mlp_input = MLP([in_channels, dim_model[0]])
self.transformer_input = TransformerBlock(in_channels=dim_model[0],
out_channels=dim_model[0])
# backbone layers
self.transformers_down = torch.nn.ModuleList()
self.transition_down = torch.nn.ModuleList()
for i in range(len(dim_model) - 1):
# Add Transition Down block followed by a Transformer block
self.transition_down.append(
TransitionDown(in_channels=dim_model[i],
out_channels=dim_model[i + 1],
k=self.k))
self.transformers_down.append(
TransformerBlock(in_channels=dim_model[i + 1],
out_channels=dim_model[i + 1]))
# class score computation
self.mlp_output = Seq(Lin(dim_model[-1], 64), ReLU(), Lin(64, 64),
ReLU(), Lin(64, out_channels))
def __init__(self, cfg):
super(BasicAttentionModel, self).__init__()
if 'modules' in cfg:
self.model_config = cfg['modules']['attention_gnn']
else:
self.model_config = cfg
self.nheads = self.model_config['nheads']
# first layer increases number of features from 4 to 16
# self.attn1 = GATConv(4, 16, heads=self.nheads, concat=False)
# first layer increases number of features from 15 to 16
self.attn1 = GATConv(16, 16, heads=self.nheads, concat=False)
# second layer increases number of features from 16 to 32
self.attn2 = GATConv(16, 32, heads=self.nheads, concat=False)
# third layer increases number of features from 32 to 64
self.attn3 = GATConv(32, 64, heads=self.nheads, concat=False)
# final prediction layer
self.edge_pred_mlp = Seq(Lin(138, 64), Dropout(p=0.2), LeakyReLU(0.12),
Dropout(p=0.2), Lin(64, 16), LeakyReLU(0.12),
Lin(16, 1), Sigmoid())
def edge_pred_model(source, target, edge_attr, u, batch):
out = torch.cat([source, target, edge_attr], dim=1)
out = self.edge_pred_mlp(out)
return out
self.edge_predictor = MetaLayer(edge_pred_model, None, None)
def __init__(self,
class_count,
nfeatures,
batch_size,
bn_momentum=0.1,
nPoints=1024):
super(PointNetInputEnhanced, self).__init__()
self.nPoints = nPoints
self.nFeatures = nfeatures + 4 # 4 is number of radii used to count the points
self.batch_size = batch_size
self.de_layer = AddNeightboursCount(
max_points=[1024, 1024, 1024, 1024], radii=[0.1, 0.2, 0.4, 0.8])
self.net = Sequential(Conv1d(self.nFeatures, 64, 1, stride=1), ReLU(),
Conv1d(64, 64, 1, stride=1), ReLU(),
Conv1d(64, 128, 1, stride=1), ReLU(),
Conv1d(128, 1024, 1, stride=1), ReLU())
self.sa3_module = GlobalSAModule() # Maxpool
# Classification Layers
self.lin1 = Lin(1024, 512)
self.bn1 = BatchNorm1d(512, momentum=bn_momentum)
self.lin2 = Lin(512, 256)
self.bn2 = BatchNorm1d(256, momentum=bn_momentum)
self.lin3 = Lin(256, class_count)
def __init__(self, class_count, nfeatures=3):
super(PointNet2MSGClass, self).__init__()
self.sa1_module = SAModuleMSG(512, [0.1, 0.2, 0.4], [16, 32, 128], [
MLP([nfeatures, 32, 32, 64]),
MLP([nfeatures, 64, 64, 128]),
MLP([nfeatures, 64, 96, 128])
])
#Because we concat the out of each layer as a feature of each point
nFeaturesL2 = 3 + 64 + 128 + 128
self.sa2_module = SAModuleMSG(128, [0.2, 0.4, 0.8], [32, 64, 128], [
MLP([nFeaturesL2, 64, 64, 128]),
MLP([nFeaturesL2, 128, 128, 256]),
MLP([nFeaturesL2, 128, 128, 256])
])
nFeaturesL3 = 3 + 128 + 256 + 256
self.sa3_module = GlobalSAModule(MLP([nFeaturesL3, 256, 512, 1024]))
#Classification Layers
self.lin1 = Lin(1024, 512)
self.bn1 = nn.BatchNorm1d(512)
self.lin2 = Lin(512, 256)
self.bn2 = nn.BatchNorm1d(256)
self.lin3 = Lin(256, class_count)
def __init__(self,
in_channels,
hidden_channels,
out_channels,
num_layers,
heads,
dropout=0.5):
super(GAT, self).__init__()
self.num_layers = num_layers
self.dropout = dropout
self.convs = torch.nn.ModuleList()
self.convs.append(GATConv(dataset.num_features, hidden_channels,
heads))
for _ in range(num_layers - 2):
self.convs.append(
GATConv(heads * hidden_channels, hidden_channels, heads))
self.convs.append(
GATConv(heads * hidden_channels, out_channels, heads,
concat=False))
self.skips = torch.nn.ModuleList()
self.skips.append(Lin(dataset.num_features, hidden_channels * heads))
for _ in range(num_layers - 2):
self.skips.append(
Lin(hidden_channels * heads, hidden_channels * heads))
self.skips.append(Lin(hidden_channels * heads, out_channels))
def __init__(self, hidden_channels):
super(GINConv, self).__init__(aggr='mean')
self.mlp = Seq(
Lin(hidden_channels, 2 * hidden_channels), ReLU(),
Lin(2 * hidden_channels, hidden_channels))
self.eps = torch.nn.Parameter(torch.Tensor([0.]))
def __init__(self):
super(SaNet, self).__init__()
self.sa1_module = SAModule(0.25, 0.2, MLP([3 + 3, 64, 64, 128]))
self.sa2_module = SAModule(0.5, 0.4, MLP([128 + 3, 128, 128, 256]))
self.sa3_module = GlobalSAModule(MLP([256 + 3, 256, 512, 512]))
self.skip_attn1 = SkipAttention(MLP([512 + 2, 128]), MLP([256, 128]),
MLP([256, 512 + 2]),
MLP([512 + 2, 512]))
self.skip_attn2 = SkipAttention(MLP([256, 64]), MLP([128, 64]),
MLP([128, 256]), MLP([256, 256]))
self.folding1 = FoldingBlock(64, 256, [
MLP([512 + 512, 256]),
MLP([512 + 512, 256]),
MLP([512 + 512, 512 + 512]),
MLP([512 + 512, 512, 256])
], [512 + 2, 512], [1024, 512])
self.folding2 = FoldingBlock(256, 512, [
MLP([256 + 256, 64]),
MLP([256 + 256, 64]),
MLP([256 + 256, 256 + 256]),
MLP([256 + 256, 256, 128])
], [256 + 2, 256], [256, 256])
self.folding3 = FoldingBlock(512, 2048, [
MLP([128 + 128, 64]),
MLP([128 + 128, 64]),
MLP([128 + 128, 128 + 128]),
MLP([128 + 128, 128])
], [128 + 2, 128], [512, 256, 128])
self.lin = Seq(Lin(128, 64), ReLU(), Lin(64, 3))
def test_point_conv():
in_channels, out_channels = (16, 32)
edge_index = torch.tensor([[0, 0, 0, 1, 2, 3], [1, 2, 3, 0, 0, 0]])
num_nodes = edge_index.max().item() + 1
x = torch.randn((num_nodes, in_channels))
pos = torch.rand((num_nodes, 3))
norm = torch.nn.functional.normalize(torch.rand((num_nodes, 3)), dim=1)
local_nn = Seq(Lin(in_channels + 4, 32), ReLU(), Lin(32, out_channels))
global_nn = Seq(Lin(out_channels, out_channels))
conv = PPFConv(local_nn, global_nn)
assert conv.__repr__() == (
'PPFConv(local_nn=Sequential(\n'
' (0): Linear(in_features=20, out_features=32, bias=True)\n'
' (1): ReLU()\n'
' (2): Linear(in_features=32, out_features=32, bias=True)\n'
'), global_nn=Sequential(\n'
' (0): Linear(in_features=32, out_features=32, bias=True)\n'
'))')
out = conv(x, pos, norm, edge_index)
assert out.size() == (num_nodes, out_channels)
jit_conv = conv.jittable(x=x, pos=pos, norm=norm, edge_index=edge_index)
jit_conv = torch.jit.script(jit_conv)
assert jit_conv(x, pos, norm, edge_index).tolist() == out.tolist()
def __init__(self, node_size, input_feature, num_classes):
super(NetGAT, self).__init__()
self.node_per_graph = node_size
hidden_size = 256
gat_head = 8
head_size = hidden_size // gat_head
self.input_feature = input_feature
# self.linprev = MLP([input_feature, 64, 64, 64])
self.linprev = EdgeConv(MLP([input_feature * 2, 64, 64, 64]),
aggr='max')
self.conv1 = GATConv(64, head_size, gat_head)
self.bn1 = torch.nn.BatchNorm1d(hidden_size)
self.lin1 = torch.nn.Linear(64, hidden_size)
self.conv2 = GATConv(hidden_size, head_size, gat_head)
self.bn2 = torch.nn.BatchNorm1d(hidden_size)
self.lin2 = torch.nn.Linear(hidden_size, hidden_size)
self.conv3 = GATConv(hidden_size, head_size, gat_head)
self.bn3 = torch.nn.BatchNorm1d(hidden_size)
self.lin3 = torch.nn.Linear(hidden_size, hidden_size)
self.conv4 = GATConv(hidden_size, head_size, gat_head)
self.bn4 = torch.nn.BatchNorm1d(hidden_size)
self.lin4 = torch.nn.Linear(hidden_size, hidden_size)
self.mlp = Seq(Lin(2048, 512), Dropout(0.4), Lin(512, 256),
Dropout(0.4), Lin(256, num_classes))
def __init__(self):
super(EdgeModel_ONE, self).__init__()
hidden = HIDDEN_EDGE_ONE
in_channels = ENCODING_EDGE_1 + 2 * ENCODING_NODE_1
self.edge_mlp = Seq(Lin(in_channels, hidden), LeakyReLU(),
LayerNorm(hidden),
Lin(hidden, ENCODING_EDGE_1)).apply(init_weights)
def __init__(self, NUM_CLASS):
super(RotationEstimateNN, self).__init__()
self.NUM_CLASS = NUM_CLASS
self.cls_model = Classifier(NUM_CLASS)
self.cls_model.load_state_dict(
torch.load(os.path.join(BASE_DIR, 'models', 'classifier_1024',
'117_0.98_1024'),
map_location='cpu'))
# self.cls_model.load_state_dict(torch.load(os.path.join(BASE_DIR, 'models', 'classifier_1024', '117_0.98_1024')))
self.cls_model.eval()
self.pcnn1 = XConv(0, 48, dim=3, kernel_size=8, hidden_channels=32)
# in_c , out_c , nei_c , spread, n_rep
self.pcnn2 = XConv(48,
96,
dim=3,
kernel_size=12,
hidden_channels=64,
dilation=2)
self.pcnn3 = XConv(96,
192,
dim=3,
kernel_size=16,
hidden_channels=128,
dilation=2)
# self.pcnn4 = XConv(
# 192, 384, dim=3, kernel_size=16, hidden_channels=256, dilation=2)
# self.pcnn5 = XConv(
# 384, 768, dim=3, kernel_size=16, hidden_channels=512, dilation=2)
self.lin1 = Lin(200, 256)
self.lin2 = Lin(256, 128)
self.lin3 = Lin(128, 3)
def test_nn_conv():
x1 = torch.randn(4, 8)
x2 = torch.randn(2, 16)
edge_index = torch.tensor([[0, 1, 2, 3], [0, 0, 1, 1]])
row, col = edge_index
value = torch.rand(row.size(0), 3)
adj = SparseTensor(row=row, col=col, value=value, sparse_sizes=(4, 4))
nn = Seq(Lin(3, 32), ReLU(), Lin(32, 8 * 32))
conv = NNConv(8, 32, nn=nn)
assert conv.__repr__() == (
'NNConv(8, 32, aggr=add, nn=Sequential(\n'
' (0): Linear(in_features=3, out_features=32, bias=True)\n'
' (1): ReLU()\n'
' (2): Linear(in_features=32, out_features=256, bias=True)\n'
'))')
out = conv(x1, edge_index, value)
assert out.size() == (4, 32)
assert conv(x1, edge_index, value, size=(4, 4)).tolist() == out.tolist()
assert conv(x1, adj.t()).tolist() == out.tolist()
if is_full_test():
t = '(Tensor, Tensor, OptTensor, Size) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit(x1, edge_index, value).tolist() == out.tolist()
assert jit(x1, edge_index, value, size=(4, 4)).tolist() == out.tolist()
t = '(Tensor, SparseTensor, OptTensor, Size) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit(x1, adj.t()).tolist() == out.tolist()
adj = adj.sparse_resize((4, 2))
conv = NNConv((8, 16), 32, nn=nn)
assert conv.__repr__() == (
'NNConv((8, 16), 32, aggr=add, nn=Sequential(\n'
' (0): Linear(in_features=3, out_features=32, bias=True)\n'
' (1): ReLU()\n'
' (2): Linear(in_features=32, out_features=256, bias=True)\n'
'))')
out1 = conv((x1, x2), edge_index, value)
out2 = conv((x1, None), edge_index, value, (4, 2))
assert out1.size() == (2, 32)
assert out2.size() == (2, 32)
assert conv((x1, x2), edge_index, value, (4, 2)).tolist() == out1.tolist()
assert conv((x1, x2), adj.t()).tolist() == out1.tolist()
assert conv((x1, None), adj.t()).tolist() == out2.tolist()
if is_full_test():
t = '(OptPairTensor, Tensor, OptTensor, Size) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit((x1, x2), edge_index, value).tolist() == out1.tolist()
assert jit((x1, x2), edge_index, value,
size=(4, 2)).tolist() == out1.tolist()
assert jit((x1, None), edge_index, value,
size=(4, 2)).tolist() == out2.tolist()
t = '(OptPairTensor, SparseTensor, OptTensor, Size) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit((x1, x2), adj.t()).tolist() == out1.tolist()
assert jit((x1, None), adj.t()).tolist() == out2.tolist()
def __init__(self):
super(EdgeModel_TWO, self).__init__()
hidden = HIDDEN_EDGE_TWO
in_channels = NO_EDGE_FEATURES_TWO+2*NO_NODE_FEATURES_TWO
self.edge_mlp = Seq(Lin(in_channels, hidden), LeakyReLU(), LayerNorm(hidden),
Lin(hidden, hidden), LeakyReLU(), LayerNorm(hidden),
Lin(hidden, NO_EDGE_FEATURES_TWO)).apply(init_weights)
def test_dynamic_edge_conv_conv():
x1 = torch.randn(8, 16)
x2 = torch.randn(4, 16)
batch1 = torch.tensor([0, 0, 0, 0, 1, 1, 1, 1])
batch2 = torch.tensor([0, 0, 1, 1])
nn = Seq(Lin(32, 16), ReLU(), Lin(16, 32))
conv = DynamicEdgeConv(nn, k=2)
assert conv.__repr__() == (
'DynamicEdgeConv(nn=Sequential(\n'
' (0): Linear(in_features=32, out_features=16, bias=True)\n'
' (1): ReLU()\n'
' (2): Linear(in_features=16, out_features=32, bias=True)\n'
'), k=2)')
out11 = conv(x1)
assert out11.size() == (8, 32)
out12 = conv(x1, batch1)
assert out12.size() == (8, 32)
out21 = conv((x1, x2))
assert out21.size() == (4, 32)
out22 = conv((x1, x2), (batch1, batch2))
assert out22.size() == (4, 32)
t = '(Tensor, OptTensor) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit(x1).tolist() == out11.tolist()
assert jit(x1, batch1).tolist() == out12.tolist()
t = '(PairTensor, Optional[PairTensor]) -> Tensor'
jit = torch.jit.script(conv.jittable(t))
assert jit((x1, x2)).tolist() == out21.tolist()
assert jit((x1, x2), (batch1, batch2)).tolist() == out22.tolist()
def get_mlp(
in_size,
out_size,
n_hidden,
hidden_size,
activation=nn.LeakyReLU,
activate_last=True,
layer_norm=True,
):
arch = []
l_in = in_size
for l_idx in range(n_hidden):
arch.append(Lin(l_in, hidden_size))
arch.append(activation())
l_in = hidden_size
arch.append(Lin(l_in, out_size))
if activate_last:
arch.append(activation())
if layer_norm:
arch.append(LayerNorm(out_size))
return Seq(*arch)
def __init__(self, NUM_CLASS):
super(Classifier, self).__init__()
self.pcnn1 = XConv(0, 48, dim=3, kernel_size=8, hidden_channels=32)
# in_c , out_c , nei_c , spread, n_rep
self.pcnn2 = XConv(48,
96,
dim=3,
kernel_size=12,
hidden_channels=64,
dilation=2)
self.pcnn3 = XConv(96,
192,
dim=3,
kernel_size=16,
hidden_channels=128,
dilation=2)
# self.pcnn4 = XConv(
# 192, 384, dim=3, kernel_size=16, hidden_channels=256, dilation=2)
# self.pcnn5 = XConv(
# 384, 768, dim=3, kernel_size=16, hidden_channels=512, dilation=2)
self.lin1 = Lin(192, 256)
self.lin2 = Lin(256, 128)
self.lin3 = Lin(128, NUM_CLASS)
def __init__(self, class_count, n_feature=3, sort_pool_k=32):
super(PointNet2MSGSortPoolClass, self).__init__()
self.sa1_module = SAModuleMSG(512, [0.1, 0.2, 0.4], [16, 32, 128], [
MLP([n_feature, 32, 32, 64]),
MLP([n_feature, 64, 64, 128]),
MLP([n_feature, 64, 96, 128])
])
#Because we concat the outout of each layer as a feature of each point
n_features_l2 = 3 + 64 + 128 + 128
self.sa2_module = SAModuleMSG(128, [0.2, 0.4, 0.8], [32, 64, 128], [
MLP([n_features_l2, 64, 64, 128]),
MLP([n_features_l2, 128, 128, 256]),
MLP([n_features_l2, 128, 128, 256])
])
n_features_l3 = 3 + 128 + 256 + 256
self.sa3_module = GlobalSortPool(MLP([n_features_l3, 256, 512, 1024]),
sort_pool_k)
#Classification Layers
classification_point_feature = 1024 * sort_pool_k
self.lin1 = Lin(classification_point_feature, 512)
self.bn1 = nn.BatchNorm1d(512)
self.lin2 = Lin(512, 256)
self.bn2 = nn.BatchNorm1d(256)
self.lin3 = Lin(256, class_count)
def __init__(self,
in_channels,
out_channels,
beta = 0.3):
super().__init__()
self.in_channels = in_channels
self.out_channels = out_channels
self.beta = beta
text_feature_size = 220
self.conv1_img = GATConv(train_dataset.num_features, 512, heads= 3, dropout=0.2)
self.conv1_text = GATConv(text_feature_size, text_feature_size // 2 , heads= 3, dropout=0.2)
self.conv2_img = GATConv(512 * 3, 512, heads= 1, dropout=0.2)
self.conv2_text = GATConv(text_feature_size // 2 * 3, text_feature_size // 2, heads= 1, dropout=0.2)
self.conv3_img = GATConv(512, 20, heads = 1, dropout = 0.2)
self.conv3_text = GATConv(text_feature_size // 2, 20, heads = 1, dropout= 0.2)
self.conv4 = GATConv(40, 20, heads= 2, dropout=0.2)
self.pool1= EdgePooling(40)
#global_mean_pool
self.lin1 = Lin(40, 20)
self.lin2 = Lin(20, 2)
def __init__(self,
input_size,
embedding_size,
n_classes,
dropout=True,
k=5,
aggr='max',
pool_op='max'):
super(DECSeq, self).__init__()
# self.bn0 = BN(input_size)
# self.bn1 = BN(64)
# self.bn2 = BN(128)
self.conv1 = EdgeConv(MLP([2 * 3, 64, 64, 64], batch_norm=True), aggr)
self.conv2 = DynamicEdgeConv(MLP([2 * 64, 128], batch_norm=True), k,
aggr)
self.lin1 = MLP([128 + 64, 1024])
if pool_op == 'max':
self.pool = global_max_pool
if dropout:
self.mlp = Seq(MLP([1024, 512], batch_norm=True), Dropout(0.5),
MLP([512, 256], batch_norm=True), Dropout(0.5),
Lin(256, n_classes))
else:
self.mlp = Seq(MLP([1024, 512]), MLP([512, 256]),
Lin(256, n_classes))
def __init__(self, num_classes):
super(Net, self).__init__()
self.conv1 = XConv(0, 48, dim=3, kernel_size=8, hidden_channels=32)
self.conv2 = XConv(48,
96,
dim=3,
kernel_size=12,
hidden_channels=64,
dilation=2)
self.conv3 = XConv(96,
192,
dim=3,
kernel_size=16,
hidden_channels=128,
dilation=2)
self.conv4 = XConv(192,
384,
dim=3,
kernel_size=16,
hidden_channels=256,
dilation=2)
self.lin1 = Lin(384, 256)
self.lin2 = Lin(256, 128)
self.lin3 = Lin(128, num_classes)
def __init__(self):
super(GlobalModel_ONE, self).__init__()
hidden = HIDDEN_GRAPH_ONE
in_channels=ENCODING_NODE_1+ENCODING_EDGE_1
self.global_mlp = Seq(Lin(in_channels, hidden), LeakyReLU(),LayerNorm(hidden),
Lin(hidden, hidden), LeakyReLU(), LayerNorm(hidden),
Lin(hidden, NO_GRAPH_FEATURES_ONE)).apply(init_weights)
def __init__(self, node_dims, edge_dims, u_dims, hidden_size=32):
super().__init__()
mlp_1_input_size = node_dims + edge_dims
self.node_mlp_1 = Seq(Lin(mlp_1_input_size, hidden_size), ReLU(),
Lin(hidden_size, hidden_size))
mlp_2_input_size = node_dims + hidden_size
self.node_mlp_2 = Seq(Lin(mlp_2_input_size, hidden_size), ReLU(),
Lin(hidden_size, node_dims))
def test_static_gin_conv():
x = torch.randn(3, 4, 16)
edge_index = torch.tensor([[0, 0, 0, 1, 2, 3], [1, 2, 3, 0, 0, 0]])
nn = Seq(Lin(16, 32), ReLU(), Lin(32, 32))
conv = GINConv(nn, train_eps=True)
out = conv(x, edge_index)
assert out.size() == (3, 4, 32)
def __init__(self):
super(NodeDecoder, self).__init__()
self.node_mlp_1 = Seq(Lin(2+32, 32),
ReLU(),
Lin(32, 32))
self.node_mlp_2 = Seq(Lin(2+32+32, 32),
ReLU(),
Lin(32, 4))
def __init__(self):
super(GlobalModel_TWO, self).__init__()
hidden = HIDDEN_GRAPH_TWO
in_channels=NO_EDGE_FEATURES_TWO+NO_NODE_FEATURES_TWO
self.global_mlp = Seq(Lin(in_channels, hidden), LeakyReLU(),LayerNorm(hidden),
Lin(hidden, hidden), LeakyReLU(), LayerNorm(hidden),
Lin(hidden, NO_GRAPH_FEATURES_TWO)).apply(init_weights)
def __init__(self):
super(NodeModel, self).__init__()
self.node_mlp_1 = Seq(
Lin(num_node_features + num_edge_features, num_node_features),
ReLU(), Lin(num_node_features, num_node_features))
self.node_mlp_2 = Seq(
Lin(2 * num_node_features, num_node_features), ReLU(),
Lin(num_node_features, num_node_features))
def __init__(self):
super(NodeEncoder, self).__init__()
self.node_mlp_1 = Seq(Lin(4+32, 32),
ReLU(),
Lin(32, 32))
self.node_mlp_2 = Seq(Lin(4+32, 32),
ReLU(),
Lin(32, 2))
def __init__(self, ratio, r):
super(Net, self).__init__()
self.r = r
self.fc1 = MLP([3, 64])
self.mu_conv = MuConv(MLP([64, 64]), MLP([64, 20]))
self.sig_conv = SigmaConv(MLP([64, 64]), MLP([64, 20]))
self.fc3 = Lin(20, 1024)
self.fc4 = Lin(1024, 3072)
