GraphVite 大規模網絡表示學習，如何使用你自己的數據，以LINE/DeepWalk爲示例

GraphVite有兩種方式，一種是使用Python接口，一種是使用命令行的方式。上兩篇博客分別介紹了使用命令行和Python接口的方式。現在介紹下，

1.隨機生成一個graph，這裏使用networkx裏的工具，生成一個BA無標度的圖，並保存爲edge_list格式，

import networkx as nx
G = nx.barabasi_albert_graph(100, 2)
nx.write_edgelist(G, 'data/test.edge_list', data=False, delimiter='\t')

2.首先創建一個yaml文件，這裏採用test.yaml,內容如下。

https://download.csdn.net/download/dongfangxiaozi_/11953105

application:
  graph

resource:
  gpus: [0]
  cpu_per_gpu: 8
  dim: 128

format:
  delimiters: " \t\n"
  comment: "#"

graph:
  file_name: /home/xxx/.graphvite/test.edgelist
  as_undirected: true

build:
  optimizer:
    type: SGD
    lr: 0.025
    weight_decay: 0.005
  num_partition: auto
  num_negative: 1
  batch_size: 100000
  episode_size: 500

train:
  model: LINE
  num_epoch: 2000
  negative_weight: 5
  augmentation_step: 2
  random_walk_length: 40
  random_walk_batch_size: 100
  log_frequency: 1000

save:
  file_name: line_test.pkl

輸入文件使用你本地的路徑，這時候要把<>給略去。

輸出模型的路徑也進行修改。

修改輸入數據的delimiter，即分隔符，一般爲空格或者\t.

維度和模型，可以根據需要調整。

3.然後輸入命令graphvite run test.yaml.

evaluate可以略去。

很快就完成了，會看到輸出

model: LINE

optimizer: SGD

learning rate: 0.025, lr schedule: linear

weight decay: 0.005

#epoch: 2000, batch size: 100000

resume: no

positive reuse: 1, negative weight: 5

>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>>

Batch id: 0 / 3

loss = 0

4.這時候我們打開剛剛生成的表示學習向量

需要安裝easydict

pip install easydict

import pickle
from easydict import EasyDict

file_name = '/home/xxx/.graphvite/line_test.pkl'
pickle.load(open(file_name,  'rb'), encoding='utf-8')

會輸出：

{'context_embeddings': array([[-0.00822913, -0.00530941, -0.00695839, ...,  0.00390708,
          0.00219729, -0.00652534],
        [-0.01103182, -0.00352184, -0.0102704 , ...,  0.00561322,
          0.00405805, -0.01209745],
        [-0.01001213, -0.00452089, -0.00941589, ...,  0.00354438,
          0.00312015, -0.00927031],
        ...,
        [-0.00722677, -0.00073841, -0.00356204, ...,  0.00240901,
          0.00181343, -0.00524651],
        [-0.00879772, -0.00321459, -0.00741335, ...,  0.00506188,
          0.00146051, -0.00558406],
        [-0.00872622, -0.00296524, -0.00560616, ...,  0.00240658,
          0.00134236, -0.00807665]], dtype=float32),
 'id2name': ['0',
  ....,
  '98'],
 'vertex_embeddings': array([[ 0.01127131,  0.00733544,  0.00983142, ...,  0.00037273,
         -0.00423259,  0.00497633],
        [ 0.01100275,  0.00174217,  0.00472951, ..., -0.0049816 ,
         -0.00484521,  0.00895557],
        [ 0.01126207,  0.00034387,  0.00764804, ..., -0.00499946,
         -0.00046137,  0.00998339],
        ...,
        [ 0.00800165,  0.00556528,  0.01023772, ..., -0.00702143,
         -0.00526442,  0.00745916],
        [ 0.00895165,  0.00092912,  0.00540247, ..., -0.00289551,
          0.00077499,  0.00426355],
        [ 0.00849834, -0.00018509,  0.00910194, ..., -0.00081561,
         -0.00055885,  0.00917041]], dtype=float32)}