COMP9312_Project_Q2: Graph Convolutional Network¶
You can edit this file and add the code to implement the model.
Copyright By PowCoder代写 加微信 powcoder
For details about this project, please refer to the project specification.
Note: Make sure to sequentially run all the cells in each section, so that the intermediate variables / packages will carry over to the next cell
├─1: Installing dependencies
├─2: Import the necessary package
├─3: Example of using Data in PyG
├─4: Download Real world dataset Cora
├─5: Output some basic attribute of the dataset **You code here**
├─6: Graph Convolutional Network By PyG **You code here**
├─7: Graph Convolutional Network By ourselves **You code here**
├─8: (optional) Another two message passing methods in PyG **You code here**
Q2: Graph Neural Netwroks (10 points)¶
PyG(https://www.pyg.org/) is an extension library for PyTorch. It provides useful primitives to develop Graph Deep Learning models, including various graph neural network layers and a large number of benchmark datasets like
torch_geometric.nn
torch_geometric.data
torch_geometric.loader
torch_geometric.transforms
torch_geometric.utils
You might need to use a GPU for this Colab to run quickly.
Please click Runtime and then Change runtime type. Then set the hardware accelerator to GPU.
1: Installing dependencies¶
The standard colab environment does not provide the package of PyG and its dependencies.
We need first to intsall it by the following commonds.
Note: This cell might take a while (up to ~10 minutes) to run
import torch
print(“Torch has version {}”.format(torch.__version__))
Torch has version 1.12.0+cu113
# Install torch geometric
# Change the version of the pytorch by modifying the url at the end
# For example, the output the above cell is ‘Torch has version 1.11.0+cu113’,then the url is “***torch-1.11.0+cu113.html”
!pip install -q torch-scatter -f https://pytorch-geometric.com/whl/torch-1.12.0+cu113.html
!pip install -q torch-sparse -f https://pytorch-geometric.com/whl/torch-1.12.0+cu113.html
!pip install -q torch-geometric
|████████████████████████████████| 7.9 MB 21.6 MB/s
|████████████████████████████████| 3.5 MB 17.4 MB/s
|████████████████████████████████| 407 kB 29.8 MB/s
Building wheel for torch-geometric (setup.py) … done
2: Import the necessary package¶
import torch.nn.functional as F
from torch_geometric.data import Data
from torch_geometric.datasets import Planetoid
from torch_geometric.loader import DataLoader
from torch_geometric.nn import GCNConv
3: Example of using Data in PyG¶
In tutorial 1 and Q1 of project, we use a class SimpleGraph and DirectedWeightedGraph to store the graph.
The pyg also provide a class called ‘Data’ to store the graph.
# The connection of edge
# for a undirected graph, use two directed edge to represent it.
# edge_index is a tensor with shape of 2*edge_num
edge_index = torch.tensor(
[0, 1, 0, 2, 1, 2, 2, 3, 4, 0, 4, 3, 5, 4],
[1, 0, 2, 0, 2, 1, 3, 2, 0, 4, 3, 4, 4, 5],
# the type of data
dtype=torch.long
# the initial features of node
x = torch.tensor(
# there are 6 nodes
# each node has a feature of length 2
# instantiate
data = Data(x=x, edge_index=edge_index)
# There may be other attribute of Data, like y which is the label of each node for
# node classification
Now, we can print some infomation about the graph.
# the whole infomation about graph
print(data)
# the key in the graph
print(data.keys)
print(data.values)
# the feature of node
print(data[‘x’])
print(data.edge_index)
print(data.num_node_features)
Data(x=[6, 2], edge_index=[2, 14])
[‘edge_index’, ‘x’]
tensor([[-4, 2],
[ 7, 1]])
tensor([[0, 1, 0, 2, 1, 2, 2, 3, 4, 0, 4, 3, 5, 4],
[1, 0, 2, 0, 2, 1, 3, 2, 0, 4, 3, 4, 4, 5]])
4: Download Real world dataset Cora¶
Use a real world dataset ‘Cora’ which is often used for node classfication.
It contains one graph and the node in the graph belongs to 7 classes.
dataset = Planetoid(root=’/tmp/Cora’, name=’Cora’)
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.x
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.tx
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.allx
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.y
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.ty
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.ally
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.graph
Downloading https://github.com/kimiyoung/planetoid/raw/master/data/ind.cora.test.index
Processing…
5: Output some basic attribute of the dataset (3 points)¶
# the number of graph in the dataset
############################################################################
# TODO: output the number of graph in Cora
# (~1 line of code)
############################################################################
# the number of node of first graph in Cora
############################################################################
# TODO: output the number of node in the first graph of Cora
# (~1 line of code)
############################################################################
# the number of edges of first graph in Cora
############################################################################
# TODO: output the number of edges in the first graph of Cora
# (~1 line of code)
############################################################################
# the dimension of feature
############################################################################
# TODO: output dimension of node feature in the first graph of Cora
# (~1 line of code)
############################################################################
# the number of classes
############################################################################
# TODO: output the number of classes that node of Cora belongs to.
# (~1 line of code)
############################################################################
# check if there are isolated nodes
############################################################################
# TODO: check if there are isolated nodes in the first graph of Cora.
# (~1 line of code)
############################################################################
6: Graph Convolutional Network By PyG (3 points)¶
Please follow the figure below to implement the forward function.
Here, we use a two layer GCN model. The GCN layer is implemented by GCNConv which is provided by PyG.
class GCN(torch.nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.conv1 = GCNConv(in_channels, 16)
self.conv2 = GCNConv(16, out_channels)
def forward(self, data):
x, edge_index = data.x, data.edge_index
############# Your code here ############
## 1. Construct the network as shown in the figure
## 2. torch.nn.functional.relu and torch.nn.functional.dropout are useful
## For more information please refer to the documentation:
## https://pytorch.org/docs/stable/nn.functional.html
## 3. Don’t forget to set F.dropout training to self.training
## (~5 lines of code)
#########################################
# start train the model
device = torch.device(‘cuda’ if torch.cuda.is_available() else ‘cpu’)
model = GCN(dataset.num_node_features, dataset.num_classes).to(device)
data = dataset[0].to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
model.train()
for epoch in range(20):
optimizer.zero_grad()
out = model(data)
loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])
loss.backward()
optimizer.step()
# start evaluate the model
model.eval()
pred = model(data).argmax(dim=1)
correct = (pred[data.test_mask] == data.y[data.test_mask]).sum()
acc = int(correct) / int(data.test_mask.sum())
print(f’Accuracy: {acc:.4f}’)
Accuracy: 0.7880
7: Graph Convolutional Networks By ourselves (4 points)¶
There are many variants of GCN formulation. Here, we use the matrix formulation of GCN as (https://openreview.net/pdf?id=SJU4ayYgl):
\begin{equation}
X^{k+1} = \sigma(\hat{D}^{-\frac{1}{2}}\hat{A}\hat{D}^{-\frac{1}{2}}X^{k}\theta)
\end{equation}here $\sigma$ is the non-linear function. X is the feature matrix. $\theta$ is the model parameters. $D\in R^{n*n}$ is the degree matrix. $I\in R^{n*n}$ is the identity matrix. $A\in R^{n*n}$ is the adjacent matrix. $\hat{D} = D+I$. And $\hat{A} = A+I$.
class Encoder(torch.nn.Module):
def __init__(self, in_channels, hidden_channels):
super(Encoder, self).__init__()
self.in_channels = in_channels
self.hidden_channels = hidden_channels
self.linear1 = torch.nn.Linear(in_channels, hidden_channels)
def forward(self, x, edge_index):
############# Your code here ############
## 1. Construct adjacnet matrix A of the graph
## 2. Calculte the degree matrix D
## 3: Calculte A_hat
## 4: Calculte the noramlized A_hat
## 5. multiply noramlized A_hat with x and theta
## np.diag, np.power, np.eye are useful
## (~11 lines of code)
#########################################
class GCN(torch.nn.Module):
def __init__(self, in_channels, out_channels):
super().__init__()
self.conv1 = Encoder(in_channels, 16)
self.conv2 = Encoder(16, out_channels)
def forward(self, data):
x, edge_index = data.x, data.edge_index
# !!! notice here
# move your code in part 6 here.
# start train the model
device = torch.device(‘cuda’ if torch.cuda.is_available() else ‘cpu’)
model = GCN(dataset.num_node_features, dataset.num_classes).to(device)
data = dataset[0].to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
model.train()
for epoch in range(20):
optimizer.zero_grad()
out = model(data)
loss = F.nll_loss(out[data.train_mask], data.y[data.train_mask])
loss.backward()
optimizer.step()
# start evaluate the model
model.eval()
pred = model(data).argmax(dim=1)
correct = (pred[data.test_mask] == data.y[data.test_mask]).sum()
acc = int(correct) / int(data.test_mask.sum())
print(f’Accuracy: {acc:.4f}’)
Accuracy: 0.7730
8: (Optional) Use another two methods of the message passing layer in the above GNN model.(2 points)¶
We’ve defined a GNN class that uses GCNConvolutional layers, suggest and test th accuracy of two other variations of this GNN class that
uses a different message passing mechanism by replacing the above GCNConv layer.
You can use the function provided by PyG or implemented by yourself.
Include a short description about the message pass layer you choose and their accuracy. You can write them in an seperated report or you can directly edit here.
Up to 2 points for this optional question and up to 10 points for the whole Q2.
程序代写 CS代考 加微信: powcoder QQ: 1823890830 Email: powcoder@163.com