WaExt: Weight-aware Extension & Tasks for Evaluating Knowledge Graph Embeddings

WaExt is a toolset built on PyTorch and PyKEEN, designed for weight-aware evaluation and extension of Knowledge Graph Embedding (KGE) models. The toolkit provides a framework for weight-aware tasks that can extend and assess KGE models. Currently, the toolkit runs somewhat slowly, and we are working on optimizing the code to improve performance.

Dependencies

1. PyKEEN, version=1.8.1
2. PyTorch, version=1.10

Usage

To train and evaluate the model, use the following command:

python trains/train_walp_v2.py --base BASE --mode MODE --step EVALUATION_FREQUENCY --model MODEL_NAME --dataset DATASET_NAME --n_weight WEIGHT_OF_NEGATIVE_SAMPLES --dpct PERCENTAGE_OF_DATASET --epoch EPOCHES --train_batch TRAIN_BATCH_SIZE --eval_batch EVALUATION_BATCH_SIZE

• --base_mode MODE: Base of the exponential function
• --MODE: Mode of activation function
• --step: Frequency of evaluation
• --model: Name of the model
• --dataset: Name of the dataset
• --n_weight: Weight for negative samples
• --dpct: Percentage of the dataset
• --epoch: Total number of training epochs
• --train_batch: Batch size for training
• --eval_batch: Batch size for evaluation

TODO

1. Standardize variable names and add comments to key sections of the code for better readability and usability
2. Perform extensive testing to eliminate potential bugs
3. Implement a more efficient version to improve runtime performance