TensorFLow or PyTorch? Both!

• GraphGallery
• 👀 What's important
• 🚀 Installation
• 🤖 Implementations
• ⚡ Quick Start
  • Datasets
    • Planetoid
    • NPZDataset
  • Framework-neutral Tensor
  • Example of GCN model
  • Customization
  • Visualization
  • Using TensorFlow/PyTorch Backend
    • GCN using PyTorch backend
• ❓ How to add your datasets
• ❓ How to define your models
  • GCN using PyG backend
• ⭐ Road Map
• 😘 Acknowledgement

GraphGallery

GraphGallery is a gallery for benchmarking Graph Neural Networks (GNNs) with TensorFlow 2.x and PyTorch backend. GraphGallery 0.6.x is a total re-write from previous versions, and some things have changed.

NEWS:

• PyG backend and DGL backend now are available in GraphGallery
• GraphGallery now supports Multiple Graph for different tasks

👀 What's important

Differences between GraphGallery and Pytorch Geometric (PyG), Deep Graph Library (DGL), etc...

• PyG and DGL are just like TensorFlow while GraphGallery is more like Keras
• GraphGallery is a plug-and-play and user-friendly toolbox
• GraphGallery has high scalaibility for researchers and developers to use

🚀 Installation

• Build from source (latest version)

git clone https://github.com/EdisonLeeeee/GraphGallery.git
cd GraphGallery
python setup.py install

• Or using pip (stable version)

pip install -U graphgallery

GraphGallery has been tested on:

• CUDA 10.1
• TensorFlow 2.1~2.3, 2.4 is unavailable now and 2.1.2 is recommended.
• PyTorch 1.5~1.7
• Pytorch Geometric (PyG) 1.6.1
• DGL 0.5.2, 0.5.3

🤖 Implementations

Please refer to the examples directory.

⚡ Quick Start

Datasets

more details please refer to GraphData.

Planetoid

fixed datasets

from graphgallery.datasets import Planetoid
# set `verbose=False` to avoid additional outputs 
data = Planetoid('cora', verbose=False)
graph = data.graph
# here `splits` is a dict like instance
splits = data.split_nodes()
# splits.train_nodes:  training indices: 1D Numpy array
# splits.val_nodes:  validation indices: 1D Numpy array
# splits.test_nodes:  testing indices: 1D Numpy array
>>> graph
Graph(adj_matrix(2708, 2708),
      node_attr(2708, 1433),
      node_label(2708,),
      metadata=None, multiple=False)

currently the available datasets are:

>>> data.available_datasets()
('citeseer', 'cora', 'pubmed')

NPZDataset

more scalable datasets (stored with .npz)

from graphgallery.datasets import NPZDataset;
# set `verbose=False` to avoid additional outputs
data = NPZDataset('cora', verbose=False)
graph = data.graph
# here `splits` is a dict like instance
splits = data.split_nodes(random_state=42)
>>> graph
Graph(adj_matrix(2708, 2708),
      node_attr(2708, 1433),
      node_label(2708,),
      metadata=None, multiple=False)

currently the available datasets are:

>>> data.available_datasets()
('citeseer','citeseer_full','cora','cora_ml','cora_full',
 'amazon_cs','amazon_photo','coauthor_cs','coauthor_phy', 
 'polblogs', 'pubmed', 'flickr','blogcatalog','dblp')

Framework-neutral Tensor

• Strided (dense) Tensor

>>> backend()
TensorFlow 2.1.2 Backend

>>> from graphgallery import functional as gf
>>> arr = [1, 2, 3]
>>> gf.astensor(arr)
<tf.Tensor: shape=(3,), dtype=int32, numpy=array([1, 2, 3], dtype=int32)>

• Sparse Tensor

>>> import scipy.sparse as sp
>>> sp_matrix = sp.eye(3)
>>> gf.astensor(sp_matrix)
<tensorflow.python.framework.sparse_tensor.SparseTensor at 0x7f1bbc205dd8>

• also works for PyTorch, just like

>>> from graphgallery import set_backend
>>> set_backend('torch') # torch, pytorch or th
PyTorch 1.6.0+cu101 Backend

>>> gf.astensor(arr)
tensor([1, 2, 3])

>>> gf.astensor(sp_matrix)
tensor(indices=tensor([[0, 1, 2],
                       [0, 1, 2]]),
       values=tensor([1., 1., 1.]),
       size=(3, 3), nnz=3, layout=torch.sparse_coo)

• To Numpy or Scipy sparse matrix

>>> tensor = gf.astensor(arr)
>>> gf.tensoras(tensor)
array([1, 2, 3])

>>> sp_tensor = gf.astensor(sp_matrix)
>>> gf.tensoras(sp_tensor)
<3x3 sparse matrix of type '<class 'numpy.float32'>'
    with 3 stored elements in Compressed Sparse Row format>

• Or even convert one Tensor to another

>>> tensor = gf.astensor(arr, backend="tensorflow") # or "tf" in short
>>> tensor
<tf.Tensor: shape=(3,), dtype=int64, numpy=array([1, 2, 3])>
>>> gf.tensor2tensor(tensor)
tensor([1, 2, 3])

>>> sp_tensor = gf.astensor(sp_matrix, backend="tensorflow") # set backend="tensorflow" to convert to tensorflow tensor
>>> sp_tensor
<tensorflow.python.framework.sparse_tensor.SparseTensor at 0x7efb6836a898>
>>> gf.tensor2tensor(sp_tensor)
tensor(indices=tensor([[0, 1, 2],
                       [0, 1, 2]]),
       values=tensor([1., 1., 1.]),
       size=(3, 3), nnz=3, layout=torch.sparse_coo)

Example of GCN model

from graphgallery.gallery import GCN

model = GCN(graph, attr_transform="normalize_attr", device="CPU", seed=123)
# build your GCN model with default hyper-parameters
model.build()
# train your model. here splits.train_nodes and splits.val_nodes are numpy arrays
# verbose takes 0, 1, 2, 3, 4
history = model.train(splits.train_nodes, splits.val_nodes, verbose=1, epochs=100)
# test your model
# verbose takes 0, 1, 2
results = model.test(splits.nodes, verbose=1)
print(f'Test loss {results.loss:.5}, Test accuracy {results.accuracy:.2%}')

On Cora dataset:

Training...
100/100 [==============================] - 1s 14ms/step - loss: 1.0161 - accuracy: 0.9500 - val_loss: 1.4101 - val_accuracy: 0.7740 - Dur.: 1.4180
Testing...
1/1 [==============================] - 0s 62ms/step - loss: 1.4123 - accuracy: 0.8120 - Dur.: 0.0620
Test loss 1.4123, Test accuracy 81.20%

Customization

• Build your model you can use the following statement to build your model

# one hidden layer with hidden units 32 and activation function RELU
>>> model.build(hiddens=32, activations='relu')

# two hidden layer with hidden units 32, 64 and all activation functions are RELU
>>> model.build(hiddens=[32, 64], activations='relu')

# two hidden layer with hidden units 32, 64 and activation functions RELU and ELU
>>> model.build(hiddens=[32, 64], activations=['relu', 'elu'])

• Train your model

# train with validation
>>> history = model.train(splits.train_nodes, splits.val_nodes, verbose=1, epochs=100)
# train without validation
>>> history = model.train(splits.train_nodes, verbose=1, epochs=100)

here history is a tensorflow History instance.

• Test you model

>>> results = model.test(splits.test_nodes, verbose=1)
Testing...
1/1 [==============================] - 0s 62ms/step - loss: 1.4123 - accuracy: 0.8120 - Dur.: 0.0620
>>> print(f'Test loss {results.loss:.5}, Test accuracy {results.accuracy:.2%}')
Test loss 1.4123, Test accuracy 81.20%

Visualization

NOTE: you must install SciencePlots package for a better preview.

import matplotlib.pyplot as plt
with plt.style.context(['science', 'no-latex']):
    fig, axes = plt.subplots(1, 2, figsize=(15, 5))
    axes[0].plot(history.history['accuracy'], label='Training accuracy', linewidth=3)
    axes[0].plot(history.history['val_accuracyuracy'], label='Validation accuracy', linewidth=3)
    axes[0].legend(fontsize=20)
    axes[0].set_title('Accuracy', fontsize=20)
    axes[0].set_xlabel('Epochs', fontsize=20)
    axes[0].set_ylabel('Accuracy', fontsize=20)

    axes[1].plot(history.history['loss'], label='Training loss', linewidth=3)
    axes[1].plot(history.history['val_loss'], label='Validation loss', linewidth=3)
    axes[1].legend(fontsize=20)
    axes[1].set_title('Loss', fontsize=20)
    axes[1].set_xlabel('Epochs', fontsize=20)
    axes[1].set_ylabel('Loss', fontsize=20)

    plt.autoscale(tight=True)
    plt.show()        

Using TensorFlow/PyTorch Backend

>>> import graphgallery
>>> graphgallery.backend()
TensorFlow 2.1.2 Backend

>>> graphgallery.set_backend("pytorch")
PyTorch 1.6.0+cu101 Backend

GCN using PyTorch backend

# The following codes are the same with TensorFlow Backend
>>> from graphgallery.gallery import GCN
>>> model = GCN(graph, attr_transform="normalize_attr", device="GPU", seed=123);
>>> model.build()
>>> history = model.train(splits.train_nodes, splits.val_nodes, verbose=1, epochs=100)
Training...
100/100 [==============================] - 0s 5ms/step - loss: 0.6813 - accuracy: 0.9214 - val_loss: 1.0506 - val_accuracy: 0.7820 - Dur.: 0.4734
>>> results = model.test(splits.test_nodes, verbose=1)
Testing...
1/1 [==============================] - 0s 1ms/step - loss: 1.0131 - accuracy: 0.8220 - Dur.: 0.0013
>>> print(f'Test loss {results.loss:.5}, Test accuracy {results.accuracy:.2%}')
Test loss 1.0131, Test accuracy 82.20%

❓ How to add your datasets

This is motivated by gnn-benchmark

from graphgallery.data import Graph

# Load the adjacency matrix A, attribute matrix X and labels vector y
# A - scipy.sparse.csr_matrix of shape [num_nodes, num_nodes]
# X - scipy.sparse.csr_matrix or np.ndarray of shape [num_nodes, num_attrs]
# y - np.ndarray of shape [num_nodes]

mydataset = Graph(adj_matrix=A, node_attr=X, node_label=y)
# save dataset
mydataset.to_npz('path/to/mydataset.npz')
# load dataset
mydataset = Graph.from_npz('path/to/mydataset.npz')

❓ How to define your models

You can follow the codes in the folder graphgallery.gallery and write you models based on:

• TensorFlow
• PyTorch
• PyTorch Geometric (PyG)
• Deep Graph Library (DGL)

NOTE: PyG backend and DGL backend now are supported in GraphGallery!

>>> import graphgallery
>>> graphgallery.set_backend("pyg")
PyTorch Geometric 1.6.1 (PyTorch 1.6.0+cu101) Backend

GCN using PyG backend

# The following codes are the same with TensorFlow or PyTorch Backend
>>> from graphgallery.gallery import GCN
>>> model = GCN(graph, attr_transform="normalize_attr", device="GPU", seed=123);
>>> model.build()
>>> history = model.train(splits.train_nodes, splits.val_nodes, verbose=1, epochs=100)
Training...
100/100 [==============================] - 0s 3ms/step - loss: 0.5325 - accuracy: 0.9643 - val_loss: 1.0034 - val_accuracy: 0.7980 - Dur.: 0.3101
>>> results = model.test(splits.test_nodes, verbose=1)
Testing...
1/1 [==============================] - 0s 834us/step - loss: 0.9733 - accuracy: 0.8130 - Dur.: 8.2737e-04
>>> print(f'Test loss {results.loss:.5}, Test accuracy {results.accuracy:.2%}')
Test loss 0.97332, Test accuracy 81.30%

similarly, you can use DGL backend just by:

# DGL PyTorch backend
>>> graphgallery.set_backend("dgl")
# DGL TensorFlow backend
>>> graphgallery.set_backend("dgl-tf")

⭐ Road Map

• Add PyTorch models support
• Add other frameworks (PyG and DGL) support
• Add more GNN models (TF and Torch backend)
• Support for more tasks, e.g., graph Classification and link prediction
• Support for more types of graphs, e.g., Heterogeneous graph
• Add Docstrings and Documentation (Building)
• Comprehensive tutorials

😘 Acknowledgement

This project is motivated by Pytorch Geometric, Tensorflow Geometric, Stellargraph and DGL, etc., and the original implementations of the authors, thanks for their excellent works!