Link to general R GSoC 2013 page

Summary: hyperSpec is a package for working with spectra (like NMR, Infrared, Raman, Fluorescence, UV/VIS, X-ray diffraction) Links: http://hyperSpec.r-forge.r-project.org and hyperspec

Description: Areas of future development are:

• parallelization & speed-up
• file and data base import filters
• plotting
  • port to ggplot2 (basic functionality exists but needs to be extended)
  • additional plots
• specialized functions functions for working with spectra (stand alone tasks)
• GUI
  • widgets for building GUIs for spectroscopic tasks
• interface functions for conversion between hyperSpec and
  • ChemoSpec

Skills required: R programming, possibly C / C++
Basic knowledge of spectroscopy or related chemistry/physics is helpful.

Test: Check the code out from the svn repository at r-forge and contribute a function (including documentation & example) for one of the “stand-alone” tasks below.
hyperSpec has a bunch of vignettes that introduce you to the ideas & use of this package.

Feel free to contact me for further information / literature (also for a manuscript draft describing the package).

Mentor: Claudia Beleites Co-Mentors: Simon Fuller

— Claudia Beleites 2013/04/07

• many calculations can easily be parallelized with parallel versions of apply, sweep, aggregate & Co.
• modify apply so that rowSum & Co. can easily be used
• some pre-processing methods should be suitable for GPU processing at 32bit precision
• identify bottleneck functions and replace by C/C++
• maybe data.table would be a fast alternative to the data.frames

Optmized BLAS are easily used in R and the parallel package is now part of core R. Key for successful parallelization of hyperSpec’s functions is a good strategy to decide

• on which level to parallelize and to
• determine packet size

• requires C/C++
• spe (binary format, documentation available)
• Witec (see also http://gwyddion.net)
• http://www.unipress.waw.pl/fityk/xylib/
• jaz
• ...

* speed up `plotmap`

• many spatially resolved hyperSpec data sets are on a square grid, however the `levelplot`ting of such maps doesn’t make any use of this and is very slow for large objects.

• much nicer results
• user can customize the plot object / add other things afterwards.
  • users are scared of writing lattice panel functions
  • grouping / faceting is needed
  • base plots allow adding, but axis labels etc. must be set immediately & grouping is close to impossible
  • estimate: plotspc spends probably > half of the code on handing through tons of possible settings.
• plotspc needs cutting wavelength axes & stacked plotting
• proof-of-concopt functions exist already

• ternary plots of mixtures (like for phase diagrams)

hyperSpec is thought to provide an infrastructure to work with spectra.

• It is important to be able to script data analyses and include the analysis e.g. in Sweave documents.
• Spectroscopic data analysis is also a highly visual task.

Thus, hyperSpec has a very general set-up. For now, no more than very basic locator () like interaction functions are available.

• One GUI for all spectroscopic tasks would be too complicated
• Small GUIs that are very efficient for specific tasks are better.
• hyperSpec should help development of such tasks with GUI-versions of the plotting functions
  • zoom, scroll, select & measure points
  • should be able to be called communicating with each other (hot linked)

• Acinonyx is very promising, but for the moment neither stable nor on CRAN. Therefore, the GUI development has low priority at the moment.

• However, most GUI tasks have a user interface part and a “computation” task that should be handled separately (so that the calculation can be done e.g. in batch mode on a server without graphical interface). Calculation projects can be done now, and they are listed below.

• most of these tasks are related to some stand-alone topics (see below)
• interactive spike filtering for Raman spectra
  • simple spike finding algorithm available
• align different spatially resolved measurements
  • microscope image & spectral measurements
  • this is important for the presentation of results of a data analysis
  • as well as for labelling spectra of heterogeneous samples, e.g. for classification:

	->
stained section with reference diagnosis (microscope image)		class memberships of spatially resolved measurement

• Measurement GUI: define grid with arbitrary outline
  • code partially available (Matlab)
  • outline:
    • record stage movement
    • draw outline on microscope image (automatic edge finder?)
  • grid options: square, hexagonal
  • measurement order: along x/y, comb / snake, random
• pre-processing
  • offset & baseline correction, normalization, centering, etc.
  • should produce code that can be copied into scripts / Sweave documents
  • includes a number of the stand-alone tasks
• Cluster analysis GUI
• Spectra fitting / library search GUI
  • see below

• more than one spectra matrix can be in the object (for multi-way data)
• other wavelength-related data can be attached to the columns of a spectra
• Maybe even matrix decompostition data?
• compatibility to other classes
• speed up hyperSpec by using datatables instead of data.frames?

• Fourier-transform smoothing/interpolation
  • The wavelength / frequency axis is not guaranteed to be evenly spaced. One major use of this function will be interpolating data that is not evenly spaced onto an evenly spaced axis.
• blending spectra
  • e.g. during piecewise baseline correction: do transition from one baseline to the other on the overlapping wavelength range.
  • e.g. for piecewise collected spectral ranges
• ternary plotting for mixture diagrams (ggplot2)
  • plot a dot for each spectrum
  • should optionally allow points outside the triangle
  • plot hexbin / density / contour lines for the density
• bi/trivariate coloring for plotmap
  • as if there were e.g. a red and a green fluorescence label
  • option: additive / subtractive color mixing
  • a basic implementation using ggplot2 exists
• peak/spike finder
• peak/band alignment
• peak fitting
• signal-to-noise-ratio calculations
  • single spectrum
  • repeated measurements
• matrix/array method for apply that preserves dimensions
• further methods, e.g. transform
• a file import filter, see above