文章目录
- Cora、Citeseer、Pubmed
- 以Cora为例
- 数据格式示例
- Tox21 数据集
本文分享一下图卷积网络GCN里用到的一些数据集的格式
Cora、Citeseer、Pubmed
数据集 | 来源 | #图 | #节点 | #边 | #特征 | #标签(y) |
---|---|---|---|---|---|---|
Cora | “Collective classification in network data,” AI magazine,2008 | 1 | 2708 | 5429 | 1433 | 7 |
Citeseer | “Collective classification in network data,” AI magazine,2008 | 1 | 3327 | 4732 | 3703 | 6 |
Pubmed | “Collective classification in network data,” AI magazine,2008 | 1 | 19717 | 44338 | 500 | 3 |
├── gcn
│ ├── data //图数据
│ │ ├── ind.citeseer.allx
│ │ ├── ind.citeseer.ally
│ │ ├── ind.citeseer.graph
│ │ ├── ind.citeseer.test.index
│ │ ├── ind.citeseer.tx
│ │ ├── ind.citeseer.ty
│ │ ├── ind.citeseer.x
│ │ ├── ind.citeseer.y
│ │ ├── ind.cora.allx
│ │ ├── ind.cora.ally
│ │ ├── ind.cora.graph
│ │ ├── ind.cora.test.index
│ │ ├── ind.cora.tx
│ │ ├── ind.cora.ty
│ │ ├── ind.cora.x
│ │ ├── ind.cora.y
│ │ ├── ind.pubmed.allx
│ │ ├── ind.pubmed.ally
│ │ ├── ind.pubmed.graph
│ │ ├── ind.pubmed.test.index
│ │ ├── ind.pubmed.tx
│ │ ├── ind.pubmed.ty
│ │ ├── ind.pubmed.x
│ │ └── ind.pubmed.y
│ ├── __init__.py
│ ├── inits.py //初始化的公用函数
│ ├── layers.py //GCN层定义
│ ├── metrics.py //评测指标的计算
│ ├── models.py //模型结构定义
│ ├── train.py //训练
│ └── utils.py //工具函数的定义
├── LICENCE
├── README.md
├── requirements.txt
└── setup.py
三种数据都由以下八个文件组成,存储格式类似:
ind.dataset_str.x => the feature vectors of the training instances as scipy.sparse.csr.csr_matrix object;
ind.dataset_str.tx => the feature vectors of the test instances as scipy.sparse.csr.csr_matrix object;
ind.dataset_str.allx => the feature vectors of both labeled and unlabeled training instances (a superset of ind.dataset_str.x) as scipy.sparse.csr.csr_matrix object;ind.dataset_str.y => the one-hot labels of the labeled training instances as numpy.ndarray object;
ind.dataset_str.ty => the one-hot labels of the test instances as numpy.ndarray object;
ind.dataset_str.ally => the labels for instances in ind.dataset_str.allx as numpy.ndarray object;ind.dataset_str.graph => a dict in the format {index: [index_of_neighbor_nodes]} as collections.defaultdict object;
ind.dataset_str.test.index => the indices of test instances in graph, for the inductive setting as list object.All objects above must be saved using python pickle module.以cora为例:
ind.dataset_str.x => 训练实例的特征向量,是scipy.sparse.csr.csr_matrix类对象,shape:(140, 1433)
ind.dataset_str.tx => 测试实例的特征向量,shape:(1000, 1433)
ind.dataset_str.allx => 有标签的+无无标签训练实例的特征向量,是ind.dataset_str.x的超集,shape:(1708, 1433)ind.dataset_str.y => 训练实例的标签,独热编码,numpy.ndarray类的实例,是numpy.ndarray对象,shape:(140, 7)
ind.dataset_str.ty => 测试实例的标签,独热编码,numpy.ndarray类的实例,shape:(1000, 7)
ind.dataset_str.ally => 对应于ind.dataset_str.allx的标签,独热编码,shape:(1708, 7)ind.dataset_str.graph => 图数据,collections.defaultdict类的实例,格式为 {index:[index_of_neighbor_nodes]}
ind.dataset_str.test.index => 测试实例的id,2157行上述文件必须都用python的pickle模块存储
- Semi-Supervised Classification with Graph Convolutional Networks论文中的GCN是半监督学习,因此训练数据集中有的有标签有的没有标签
以Cora为例
原始数据集链接:http://linqs.cs.umd.edu/projects/projects/lbc/
数据集划分方式:https://github.com/kimiyoung/planetoid (Zhilin Yang, William W. Cohen, Ruslan Salakhutdinov, Revisiting Semi-Supervised Learning with Graph Embeddings, ICML 2016)
Cora数据集由机器学习论文组成,是近年来图深度学习很喜欢使用的数据集。在数据集中,论文分为以下七类之一:
- 基于案例
- 遗传算法
- 神经网络
- 概率方法
- 强化学习
- 规则学习
- 理论
论文的选择方式是,在最终语料库中,每篇论文引用或被至少一篇其他论文引用。整个语料库中有2708篇论文。
在词干堵塞和去除词尾后,只剩下1433个独特的单词。文档频率小于10的所有单词都被删除。cora数据集包含1433个独特单词,所以特征是1433维。0和1描述的是每个单词在paper中是否存在。
变量data是个scipy.sparse.csr.csr_matrix,类似稀疏矩阵,输出得到的是矩阵中非0的行列坐标及值
数据格式示例
(1)--------------------------------------ind.cora.x
def load_cora():names = ['x']with open("data/ind.cora.x", 'rb') as f:if sys.version_info > (3, 0):print(f) # <_io.BufferedReader name='data/ind.cora.x'>data = pkl.load(f, encoding='latin1')print(type(data)) #<class 'scipy.sparse.csr.csr_matrix'>print(data.shape) #(140, 1433)-ind.cora.x是140行,1433列的print(data.shape[0]) #row:140print(data.shape[1]) #column:1433print(data[1])# 变量data是个scipy.sparse.csr.csr_matrix,类似稀疏矩阵,输出得到的是矩阵中非0的行列坐标及值# (0, 19) 1.0# (0, 88) 1.0# (0, 149) 1.0# (0, 212) 1.0# (0, 233) 1.0# (0, 332) 1.0# (0, 336) 1.0# (0, 359) 1.0# (0, 472) 1.0# (0, 507) 1.0# (0, 548) 1.0# ...# print(data[100][1]) #IndexError: index (1) out of rangenonzero=data.nonzero()print(nonzero) #输出非零元素对应的行坐标和列坐标
# (array([ 0, 0, 0, ..., 139, 139, 139], dtype=int32), array([ 19, 81, 146, ..., 1263, 1274, 1393], dtype=int32))# nonzero是个tupleprint(type(nonzero)) #<class 'tuple'>print(nonzero[0]) #行:[ 0 0 0 ... 139 139 139]print(nonzero[1]) #列:[ 19 81 146 ... 1263 1274 1393]print(nonzero[1][0]) #19print(data.toarray())
# [[0. 0. 0. ... 0. 0. 0.]
# [0. 0. 0. ... 0. 0. 0.]
# [0. 0. 0. ... 0. 0. 0.]
# ...
# [0. 0. 0. ... 0. 1. 0.]
# [0. 0. 0. ... 0. 0. 0.]
# [0. 1. 0. ... 0. 0. 0.]](2)--------------------------------------ind.cora.ydef load_cora():with open("data/ind.cora.y", 'rb') as f:if sys.version_info > (3, 0):print(f) #<_io.BufferedReader name='data/ind.cora.y'>data = pkl.load(f, encoding='latin1')print(type(data)) #<class 'numpy.ndarray'>print(data.shape) #(140, 7)print(data.shape[0]) #row:140print(data.shape[1]) #column:7print(data[1]) #[0 0 0 0 1 0 0](3)--------------------------------------ind.cora.graphdef load_cora():with open("data/ind.cora.graph", 'rb') as f:if sys.version_info > (3, 0):data = pkl.load(f, encoding='latin1')print(type(data)) #<class 'collections.defaultdict'>print(data)
# defaultdict(<class 'list'>, {0: [633, 1862, 2582], 1: [2, 652, 654], 2: [1986, 332, 1666, 1, 1454],
# , ... ,
# 2706: [165, 2707, 1473, 169], 2707: [598, 165, 1473, 2706]})(4)--------------------------------------ind.cora.test.indextest_idx_reorder = parse_index_file("data/ind.{}.test.index".format(dataset_str))
print("test index:",test_idx_reorder)
#test index: [2692, 2532, 2050, 1715, 2362, 2609, 2622, 1975, 2081, 1767, 2263,..]
print("min_index:",min(test_idx_reorder))
# min_index: 1708(5)citeseer数据集中一些孤立点的特殊处理#处理citeseer中一些孤立的点if dataset_str == 'citeseer':# Fix citeseer dataset (there are some isolated nodes in the graph)# Find isolated nodes, add them as zero-vecs into the right positiontest_idx_range_full = range(min(test_idx_reorder), max(test_idx_reorder)+1)# print("test_idx_range_full.length",len(test_idx_range_full))#test_idx_range_full.length 1015#转化成LIL格式的稀疏矩阵,tx_extended.shape=(1015,1433)tx_extended = sp.lil_matrix((len(test_idx_range_full), x.shape[1]))# print(tx_extended)#[2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325# ....# 3321 3322 3323 3324 3325 3326]#test_idx_range-min(test_idx_range):列表中每个元素都减去min(test_idx_range),即将test_idx_range列表中的index值变为从0开始编号tx_extended[test_idx_range-min(test_idx_range), :] = tx# print(tx_extended.shape) #(1015, 3703)# print(tx_extended)# (0, 19) 1.0# (0, 21) 1.0# (0, 169) 1.0# (0, 170) 1.0# (0, 425) 1.0# ...# (1014, 3243) 1.0# (1014, 3351) 1.0# (1014, 3472) 1.0tx = tx_extended# print(tx.shape)# (1015, 3703)#997,994,993,980,938...等15行全为0ty_extended = np.zeros((len(test_idx_range_full), y.shape[1]))ty_extended[test_idx_range-min(test_idx_range), :] = tyty = ty_extended# for i in range(ty.shape[0]):# print(i," ",ty[i])# # 980 [0. 0. 0. 0. 0. 0.]# # 994 [0. 0. 0. 0. 0. 0.]# # 993 [0. 0. 0. 0. 0. 0.]
- allx是训练集中的所有训练实例,包含有标签的和无标签的,从0-1707,共1708个
- ally是allx对应的标签,从1708-2707,共1000个
- citeseer的测试数据集中有一些孤立的点(test.index中没有对应的索引,15个),可把这些点当作特征全为0的节点加入到测练集tx中,并且对应的标签在ty中
- 输入是一张整图,因此将tx和allx拼起来作为feature
- 没有标签的数据的y值:[0,0,0,0,0,0,0]
- 数据集中的特征也是稀疏的,用LIL稀疏矩阵存储,格式如下
A=np.array([[1,0,2,0],[0,0,0,0],[3,0,0,0],[1,0,0,4]])
AS=sp.lil_matrix(A)
print(AS)
# (0, 0) 1
# (0, 2) 2
# (2, 0) 3
# (3, 0) 1
# (3, 3) 4
Tox21 数据集
此数据集来源于一个PubChem网站的一个2014年的竞赛:https://tripod.nih.gov/tox21/challenge/about.jsp
PubChem是美国国立卫生研究院(NIH)的开放化学数据库,是世界上最大的免费化学物信息集合。
PubChem的数据由数百个数据源提供,包括:政府机构,化学品供应商,期刊出版商等。
21世纪的毒理学(Tox21)计划是NIH,环境保护局和食品药品管理局的联邦合作计划,旨在开发更好的毒性评估方法。目标是快速有效地测试某些化合物是否有可能破坏人体中可能导致不良健康影响的过程。Tox21数据集是其中一个比赛用到的数据集,包含了12个毒理试验测定的化学合成物质的结构信息
- 雌激素受体α,LBD(ER,LBD)
- 雌激素受体α,full(ER,full)
- 芳香
- 芳烃受体(AhR)
- 雄激素受体,full(AR,full)
- 雄激素受体,LBD(AR,LBD)
- 过氧化物酶体增殖物激活受体γ(PPAR-γ)
- 核因子(红细胞衍生的2)样2 /抗氧化反应元件(Nrf2 / ARE)
- 热休克因子反应元件(HSE)
- ATAD5
- 线粒体膜电位(MMP)
- P53
每个毒理实验测试的都是PUBCHEM_SID从144203552-144214049共10486个化合物,包括环保化合物、一些上市药物等物质的活性结果。
例如,p53实验的测定结果可以在线查看。
- PubChem AID:生物活性鉴定记录ID
- PubChem SID:物质ID
- PubChem CID:化合物ID
数据集可在此下载:https://tripod.nih.gov/tox21/challenge/data.jsp#
训练集和测试集都是由多个分子结构构成的sdf格式的文件。
一个分子的信息存储格式如下:
- 第一行:一般作为分子名字,如NCGC00255644-01,有时为空格
- 第二行:注释,Marvin 07111412562D
- 第三行:一般是空行
- 第四行:是原子个数 键的个数等的起始行。
- M END所在行结束原子个数 键的个数等信息。
下面是属性值,属性个数不定
- 属性1
- 属性1值
- 空行
- 属性2
- 属性2值
- 空行
- (以四个美元符号结束一个分子的信息存储。)
- 在训练集中,标签为”Active“,“1”表示活性,“0”表示没有活性
- 在测试集中,没有标签”Active“
训练集中一个分子的信息存储格式如下:
NCGC00255644-01Marvin 07111412562D 26 27 0 0 1 0 999 V20004.5831 -4.3075 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 05.2840 -3.9061 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 05.9910 -4.3075 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 05.2840 -3.0973 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 01.4379 -1.6595 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.4379 -2.4863 0.0000 C 0 0 1 0 0 0 0 0 0 0 0 02.1508 -2.0609 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.4379 -3.3010 0.0000 C 0 0 2 0 0 0 0 0 0 0 0 00.7070 -2.0609 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 02.8577 -2.4863 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 02.1508 -1.2342 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.7070 -3.7084 0.0000 C 0 0 1 0 0 0 0 0 0 0 0 02.1508 -3.7084 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.0000 -2.4863 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 02.8577 -3.3010 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 03.5646 -2.0609 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 02.8577 -0.8388 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.1323 -4.4273 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.3056 -4.4273 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.0000 -3.3010 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 03.5646 -1.2342 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.7189 -5.1463 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 04.2955 -0.8388 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 05.0085 -1.2342 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 04.2955 0.0000 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.4379 -4.1338 0.0000 H 0 0 0 0 0 0 0 0 0 0 0 01 2 1 0 0 0 02 3 1 0 0 0 02 4 2 0 0 0 06 5 1 1 0 0 06 7 1 0 0 0 06 8 1 0 0 0 06 9 1 0 0 0 07 10 1 0 0 0 07 11 2 0 0 0 08 12 1 0 0 0 08 13 1 0 0 0 08 26 1 6 0 0 09 14 1 0 0 0 010 15 1 0 0 0 010 16 2 0 0 0 011 17 1 0 0 0 012 18 1 6 0 0 012 19 1 1 0 0 012 20 1 0 0 0 013 15 1 0 0 0 014 20 1 0 0 0 016 21 1 0 0 0 017 21 2 0 0 0 018 22 1 0 0 0 021 23 1 0 0 0 023 24 1 0 0 0 023 25 1 0 0 0 0
M END
> <Formula>
C22H35NO2> <FW>
345.5188 (60.0520+285.4668)> <DSSTox_CID>
27102> <Active>
0$$$$
测试集中一个分子的信息存储格式如下:
NCGC00261443Marvin 10161415332D 20 22 0 0 1 0 999 V20000.5185 2.9762 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.2330 2.5637 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 01.2330 1.7387 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 00.5185 1.3262 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0-0.2661 1.5812 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 0-0.7510 0.9137 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0-0.2661 0.2463 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 0-0.5210 -0.5383 0.0000 C 0 0 2 0 0 0 0 0 0 0 0 0-1.3056 -0.7933 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 0-1.3056 -1.6183 0.0000 C 0 0 2 0 0 0 0 0 0 0 0 0-1.9731 -2.1032 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0-2.7268 -1.7676 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 0-0.5210 -1.8732 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 0-0.2661 -2.6578 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 0-0.0361 -1.2058 0.0000 C 0 0 1 0 0 0 0 0 0 0 0 00.7889 -1.2058 0.0000 O 0 0 0 0 0 0 0 0 0 0 0 00.5185 0.5012 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.2330 0.0887 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 01.9475 0.5012 0.0000 C 0 0 0 0 0 0 0 0 0 0 0 01.9475 1.3262 0.0000 N 0 0 0 0 0 0 0 0 0 0 0 01 2 1 0 0 0 02 3 1 0 0 0 03 4 2 0 0 0 04 5 1 0 0 0 05 6 2 0 0 0 06 7 1 0 0 0 07 8 1 0 0 0 08 9 1 1 0 0 09 10 1 0 0 0 010 11 1 1 0 0 011 12 1 0 0 0 010 13 1 0 0 0 013 14 1 0 0 0 013 15 1 0 0 0 08 15 1 0 0 0 015 16 1 6 0 0 07 17 1 0 0 0 04 17 1 0 0 0 017 18 2 0 0 0 018 19 1 0 0 0 019 20 2 0 0 0 03 20 1 0 0 0 0
M END
> <Compound ID>
NCGC00261443> <Compound Batch ID>
NCGC00261443-01> <NR-AR>
0> <NR-AR-LBD>
0> <NR-AhR>
0> <NR-ER>
0> <NR-ER-LBD>
0> <NR-PPAR-gamma>
0> <SR-ARE>
0> <SR-ATAD5>
1> <SR-HSE>
0> <SR-MMP>
0> <SR-p53>
0$$$$
目标应该就是根据训练集的分子结构信息和是否是活性的标签去预测测试集中的分子结构的活性。训练集中可能是一个分子构成一张图,里面的原子和健构成节点和边,但是没有找到关于数据集中原子和健部分的数据更具体介绍,不知道每一行数据的意义。
有错误的地方还望不吝指出,欢迎进群交流GNNs&GCNs(入群备注信息!!!,格式:姓名 -(学校或其他机构信息)- 研究方向)。