Free Access

Table 2.

Summary of the different methods compared in this work.

Type Rejection
Le Phare Template-fitting Weak
CPz Random forest classification + template-fitting Weak
Phosphoros Template-fitting No
EAzY Template-fitting Strong
METAPHOR Machine-learning: neural network Strong
ANNz Machine-learning: neural network No
GPz Machine-learning: Gaussian processes Weak
GBRT Machine-learning: boosted decision trees Weak
RF Machine-learning: random forest No
Adaboost Machine-learning: boosted decision trees No
DNF Machine-learning: nearest neighbor Strong
frankenz Machine-learning: nearest neighbor Strong
NNPZ Machine-learning: nearest neighbor No

Notes. Columns are: the name of the code, the type of the approach (template-fitting or machine-learning) and whether a rejection of the results is applied or not. We qualify as strong a rejection of more than 15% of the full spectroscopic sample (more than 10 594 sources remaining; see Table B.1), otherwise it is considered to be a weak one.

Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.

Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.

Initial download of the metrics may take a while.