Description of basic parameters

num_layers : The number of layers in some GNN model.

decoder_model : The name of decoder model, in some model.

eval_task : The task of validation, default link_prediction.

calc_hits : Calculate the hit rate, default [1,3,5,10].

gpu : Select the GPU in training, default cuda:0.

filter_flag : Filter in negative sampling.

use_wandb : Use “weight and bias” to record the result.

use_weight : Use subsampling weight.

checkpoint_dir : The checkpoint model path.

save_config : Save parameters config file.

load_config : Load parameters config file.

config_path : The config path.

model_name : The name of model.

monitor : The index name of early stopping.

dataset_name: The name of dataset.

data_path : The folder path of dataset.

data_class : The name of data preprocessing module, default KGDataModule.

litmodel_name : The name of processing module of training, evaluation and testing, default KGELitModel.

train_sampler_class : Sampling method used in training, default UniSampler.

test_sampler_class : Sampling method used in validation and testing, default TestSampler.

loss_name : The name of loss function.

negative_adversarial_sampling : Use self-adversarial negative sampling.

optim_name : The name of optimizer.

seed: Random seed.

margin : The fixed margin in loss function.

adv_temp : The temperature of sampling in self-adversarial negative sampling.

emb_dim : The embedding dimension in KGE model.

out_dim : The output embedding dimmension in some KGE model.

max_epochs : The maximum epoch in training.

lr : Learning rate

train_bs : Batch size in training.

eval_bs : Bathc size in evaluation and testing.

num_neg : The number of negative samples corresponding to each positive sample

num_ent : The number of entity, autogenerate.

num_rel : The number of relation, autogenerate.

check_per_epoch : Evaluation per n epoch of training.

early_stop_patience : If the number of consecutive bad results is n, early stop.