Configuration Files

Experiment configuration

An experiment configuration starts with an application type, and contains settings for resource, format, graph, build, load, train, evaluate and save stages.

Here is the configuration used in Quick Start. quick_start.yaml

The stages are configured as follows.

application: [type]

The application type can be graph, word graph, knowledge graph or visualization.

resource:
  gpus: [list of GPU ids]
  gpu_memory_limit: [limit for each GPU in bytes]
  cpu_per_gpu: [CPU thread per GPU]
  dim: [dim]

Note

For optimal performance, modules are compiled with pre-defined dimensions in C++. As a drawback, only dimensions that are powers of 2 are supported in the library.

format:
  delimiters: [string of delimiter characters]
  comment: [prefix of comment strings]

Format section is optional. By default, delimiters are any blank character and comment is “#”, following the Python style.

graph:
  file_name: [file name]
  as_undirected: [symmetrize the graph or not]

For standard datasets, we can specify its file name by <[dataset].[split]>. This would make the configuration file independent of the path.

build:
  optimizer:
    type: [type]
    lr: [learning rate]
    weight_decay: [weight decay]
    schedule: [learning rate schedule]
    # and other optimizer-specific configuration
  num_partition: [number of partitions]
  num_negative: [number of negative samples]
  batch_size: [batch size]
  episode_size: [episode size]

The number of partitions determines how to deal with multi-GPU or large graph cases. The more partitions, the less GPU memory consumption and speed. The episode size controls the synchronization frequency across partitions.

See section 3.2 in GraphVite paper for a detailed illustration.

load:
  file_name: [file name]

Loading a model is optional.

train:
  model: [model]
  num_epoch: [number of epochs]
  resume: [resume training or not]
  log_frequency: [log frequency in batches]
  # and other application-specific configuration

To resume training from a loaded model, set resume to true in train.

See also

Training interface: Graph, Knowledge graph, Visualization

evaluate:
  - task: [task]
    # and other task-specific configuration
  - task: [task]
    ...

Evaluation is optional. There may be multiple evaluation tasks.

See also

Evaluation tasks:

• Graph: link prediction, node classification

• Knowledge graph: link prediction, entity prediction

• Visualization: visualization, animation, hierarchy

save:
    file_name: [file name]
    save_hyperparameter: [save hyperparameters or not]

Saving the model is optional.

For more detailed settings, we recommend to read the baseline configurations for concrete examples. They can be found under config/ in the Python package, or in the GitHub repository.

Global configuration

We can overwrite the global settings of GraphVite in ~/.graphvite/config.yaml.

backend: [graphvite or torch]
dataset_path: [path to store downloaded datasets]
float_type: [default float type]
index_type: [default index type]

By default, the evaluation backend is graphvite. The datasets are stored in ~/.graphvite/dataset. The data types are float32 and uint32 respectively.