The catalog files, the scripts and the functions used for “Taxonomic classification of asteroids based on MOVIS near-infrared colors” (Popescu et al. 2018, A&A)
!!!!!!! ******** The catalog file with the taxonomic classification
/Files/MOVIS-CTax.csv
-
"Files" contains all the files required to generate the taxonomic classification. 371SpectraDeMeoColor.csv - The synthetic colors computed from the 371 spectra published by DeMeo et al 2009. This represent the training set for the classification algorithms DeMeoMeanColors.csv - Average colors and standard deviation for Bus-DeMeo taxonomic classes MOVIS-CTax.csv - The results file - the final classification assigned to the MOVIS-C catalog. This file will be also accessible via CDS - Stasbourg, as described in the article MovisKnnCheck.csv, MovisProbCheck.csv - Intermediate file generated by the algorithm. These two files are merged together in the MovisProbCheck.csv file
-
"KNN" contains the KNN classification code in the Jupyter Notebook for Python code. Notice that the parameters of the algorithm, or even the algorithm can be modified in a easilly manner if you want to verify it with other methods This code is useful if you want to study an individual object and to check the similarities/probabilities with each taxonomic class Thanks to Radu Stoicescu for providing this code.
-
"Probabilistic" contains the Octave code for probabilistic classification. For an individual object, use: ReportTaxonomyMOVISCobj(). Example:ReportTaxonomyMOVISCobj('5587') It reports the probabilities for each class, as described in Popescu et al 2018, and the position in the Y-J versus J-Ks color space
!!! Disclaimer:
We recall that individual objects follow the statistics shown in Fig. 2 from the paper and their classification is made with a probability that depends on their color errors.
There are several errors not accounted by this work. For their description, please see Popescu et al. 2018, and Popescu et al. 2016
a) identification errors - improper identification of targets in the survey data. These are in the order of ~2%.
b) overlapping with a background source. These are less than 1 %;
c) Errors due to rotation/lightcurve during the time interval between the observations with individual filters. Statistically, these values translate into an uncertainty of ∼0.03 magnitudes for (Y-J) and (J-Ks), which have about a ∼7 min time interval between the observations with each filter.