next up previous
Next: Scientific Results in Press Up: IRAM Newsletter 67 (August 2006) Previous: New Radio Link between

A new CLASS directed toward large data sets

CLASS is a GILDAS (http://www.iram.fr/GILDAS) software that has been used for years at the 30m-telescope to reduce and analyse the spectroscopic data. CLASS is also used in many other facilities (e.g. APEX, CSO, HHT, Effelsberg) and it is considered for use by Herschel/HIFI.

The development of CLASS started in the 1980s, and it was therefore written in FORTRAN77 and tailored to reduce pointed observations. On-The-Fly (OTF) support was added in the 90s but it showed limitations as the quantity of OTF data increased quickly over the years. Indeed, the advent of multibeam spectral-line receivers (e.g. HERA at the 30m) has increased the number of spectra by several orders of magnitude thus demanding enhanced software capabilities for the treatment of large datasets. For instance, a fully sampled ($4''$ sampling) map of $1\deg{} \times 1\deg{}$ is done with the HERA receiver in only 30 hours ($4''$/s scanning velocity) resulting in $8\times10^5$ reduced spectra. This represents 1.6 GB (assuming 500 spectral channels per spectrum). It was thus decided to develop a new version of CLASS, written in FORTRAN90, that would enable an easy reduction of large OTF data sets without loosing, of course, all existing tools that work satisfactorily. Most of the rewriting happened in 2005 and 30m users started to use it successfully during the winter 2005-2006.

The new developments were directed toward a better handling of large data sets. The underlying idea is to enable complex operations on a large quantity of spectra while keeping an easy access to every single spectrum. Both kind of operations are optimized to require the minimum number of typed commands to obtain an efficient user interface. To do this, the same command names may be used on a single spectrum or on a collection of spectra grouped in a given index list. No assumption is made a priori on the relationship between spectra that are selected in the index: e.g. the spectra may belong to many different scans. As some operations require consistent data sets, a new command was introduced to optionally check this consistency.

Figure 2: 2D visualization of about 4300 spectra sorted by increasing lambda offset. The 2 polygons displayed in white enable to define the area that contains the signal and should therefore not be used for baseline fitting.
\begin{figure*}\begin{center}
\mbox{\psfig{file=aug06-2dplot.eps,width=13cm}}
\end{center} \end{figure*}

Figure 3: Data exploration utility using position/velocity plots with interactive cuts at constant position or velocity.
\begin{figure*}\begin{center}
\mbox{\psfig{file=aug06-exploration.eps,width=17cm}}
\end{center} \end{figure*}

Among the operations already optimized toward large data sets are listing, visualization and baseline subtraction. Listing capabilities were expanded to quickly display in a compact form the whole file content in a way similar to a table of content. Two-dimensional displays (intensities of all the spectra in the index list as a function of the spectra number and of the velocity/frequency, see Fig. 2) enable the quick inspection of the data (e.g. platforming, ripples, line strength, kinematics). This feature is further improved by the possibility to sort the spectra in different orders (by time, by coordinates, by backend name, etc...). Using those basic capabilities, more sophisticated interactive utilities as the one shown in Fig. 3 have been developed to ease the exploration of the large data sets. Baseline subtraction being a crucial operation in spectral line data reduction, we have extended the former capabilities of CLASS to enable the definition of baseline windows on an arbitrary number of spectra through the drawing of an arbitrary number of polygons on the two-dimensional displays (see Fig. 2).

Others improvements were also implemented. For instance, the building of images from non-gridded spectra was fully rewritten to enable a better control of the parameters (resolution, grid definition, etc...) by the users. Among the various analysis tools of CLASS, the fitting ones have been grouped in a new FIT language for clarity. All GREG utilities, including the facilities to visualize gridded spectra cubes (e.g. displaying channel maps or position-velocity diagrams, with interactive possibilities), are now available from CLASS. Finally, the documentation has been updated, taking into account previous lacks and new features. For the coming years, it is planned to anticipate the increase in the bandwidths of the instruments by removing CLASS limitations in the definition of the frequency/velocity axis of the spectra (i.e. definition of non-linear axis) while continuing to enlarge the capabilities of CLASS for reduction and analysis of large datasets from either single or multibeam receivers.

A technical memo ("CLASS Evolution: I. Improved OTF support") describing in details all the new features is available at

http://www.iram.fr/GENERAL/reports/class-evol1.pdf
Starting with the mar06 release of GILDAS, both flavors of CLASS (old and new) are compiled by default. They are thus directly available, after GILDAS installation, by typing class or class90 at the shell prompt.



P. HILY-BLANT, J. PETY & S. GUILLOTEAU


next up previous
Next: Scientific Results in Press Up: IRAM Newsletter 67 (August 2006) Previous: New Radio Link between