This page analyses how well our machine learning-based identifier did on our testing and training sets. The testing set is a collection of Hebeloma collections for which we have identified a species by expert and/or genetic analysis. The testing set however was not used to "train" the identifier in how to recognise species. This set can be used therefore to determine whether the identifier is working well.

The chart below gives an overview of the success of each identifier. For any given collection, the identifier ranks, in order of likelihood from highest to lowest, the chances of that collection being of a given species. The nth point on the charts shows the proportion of collections where the actual species was in the top N "guesses" made by the identifier.

Choose an individual identifier to drill-down into detailed results. A table is shown for each section or subsection. The first column shows the name of the species as determined by expert analysis. The second column shows the number of collections in the set. The third column counts the number of times the correct species was the top guess of the classifier. The fourth to seventh columns count the number of times the correct species was the 2nd, 3rd, 4th or 5th guess of the identifier respectively. The eighth column records the number of times where the correct species was not in the top five guesses.

Finally, for those collections where the correct species was NOT the top guess, we record which species *was* chosen by the identifier. In this column, species in the same subsection as the true species (or same section if there is no subsection) are highlighted in bold blue. Species in the same section but wrong subsection are highlighted in light blue. It is common for mis-classified species to at least have been classified in the correct section; this is a sign that the classifier "agrees" with the genetic and morphological decomposition of the genus into sections.

  • arrow_drop_downarrow_drop_upSummary results for testing set
  • arrow_drop_downarrow_drop_upSummary results for training set
  • arrow_drop_downarrow_drop_upFull list of incorrect collections