In the following tables are present different metrics obtained after training neural networks to classify partially symbolic structures. Case studies of varying complexity were considered as well as different scopes. Regarding the training phase, for each scope k we report:
- f : number of features, i.e., length of the vectors representing the objects
- time : training time in seconds
With respect to the evaluation phase, for the positive instances we show:
- tp : true positives (actually positive instances that are classified as positive)
- fp : false positives (actually negative instances that are classified as positive)
- precision % : percentage of instances classified as positive that are actually positive (tp/(tp+fp))
- recall % : percentage of actually positive instances classified as positive (tp/total)
And for the negative ones:
- tn : true negatives (actually negative instances that are classified as negative)
- fn : false negatives (actually positive instances that are classified as negative)
- precision % : percentage of instances classified as negative that are actually negative (tn/(tn+fn))
- recall % : percentage of actually negative instances classified as negative (tn/total)