Optimising a model
As each model comprises of separate components, we optimise each component individually. The components in question are:
- GNN (
graph
) - Classification head (
clf_head
) - Keyword pruning threshold (
threshold_optimization
)
All separate components can be optimised using the --optimize
flag and take three arguments: optimization component,
dataset, gnn_type
E.g.:
python main.py --optimize 'clf_head' 'canary' 'GAT'
The additional parameters are those set in the default settings files.
There is a manual optimization for the threshold available as well, specified with threshold_experiment
useful for
evaluating the effect of the threshold for specific ranges.