kge
kge copied to clipboard
uma-pi1
LibKGE - A knowledge graph embedding library for reproducible research
As title says, this PR adds loss/cost on validation data to address https://github.com/uma-pi1/kge/issues/2 . Reuse the relevant parts of TrainingJob to run the train job on validation data. This also...
Right now, only possible with folder structure as used in training.
Currently we support only one evaluation type during a main job's run. We should support if users want to log multiple evaluation types during training (e.g. entity ranking, training_loss and...
How about we introduce a default config that - when present - is loaded after config-default.yaml, that has some default configurations that are specific for a machine, e.g. job.device, search.devicepool,...
Right now, weighted regularization (which regularizes batch entities) scales better then unweigthed regularization (which regularizes all entities) when negative sampling is used. Whether we regularize only the batch entities or...
Right now, scorer uses about 4x memory than needed but is faster.
`valid.early_stopping.min_threshold.epochs` should be w.r.t. to validation runs and be renamed (e.g., to `valid.early_stopping.metric_value_threshold.after_validations` or so.
https://github.com/rufex2001/kge/blob/9d83e43f5085e4a0d30d70536e4c1772389907cd/kge/job/entity_ranking.py#L72
Most notably, a hyperparameter optimization package may be used to determine which configurations appear most promising and evaluate those first.