config.yaml

task_name: distributed_metapath2vec

# ---------------------------数据配置-------------------------------------------------#
# for data preprocessing
data_path: ./data/net_aminer
author_label_file: ./data/label/googlescholar.8area.author.label.txt
venue_label_file: ./data/label/googlescholar.8area.venue.label.txt
processed_path: ./graph_data

# for pgl graph engine
etype2files: "p2a:./graph_data/paper2author_edges.txt,p2c:./graph_data/paper2conf_edges.txt"
ntype2files: "p:./graph_data/node_types.txt,a:./graph_data/node_types.txt,c:./graph_data/node_types.txt"
symmetry: True
meta_path: "c2p-p2a-a2p-p2c"
first_node_type: "c"

shard_num: 100

walk_len: 24
win_size: 3
neg_num: 5
walk_times: 20


# ---------------------------模型参数配置---------------------------------------------#
model_type: SkipGramModel
warm_start_from: null
num_nodes: 5000000
embed_size: 64
sparse_embed: False

# ---------------------------训练参数配置---------------------------------------------#
epochs: 1
num_workers: 4
lr: 0.001
lazy_mode: False
batch_node_size: 200
batch_pair_size: 1000
pair_stream_shuffle_size: 100000
log_dir: ./logs
output_dir: ./outputs
save_dir: ./checkpoints
log_steps: 1000