Scaled Data

Scaled Data

This suggestion is one of the outcomes of the NeXus for Synchrotrons Workshop at PSI: http://lns00.psi.ch/nexus2010

The Suggestion

NeXus STRONGLY suggests to store data as arrays of physical values in C storage order. However, for cases where this is not possible or would cause an efficiency concern when writing allow to store raw data. Such data must be annotated with additional attributes as described below in order to allow reading software to reconstruct the true physical value.

The Reasoning

The data rates possible at synchrotron facilities and the new pixel detectors test current computing technology to their limits. There may not be enough time to scale or convert data on the fly before writing to disk. In some occasions significant space savings can be obtained by storing data as short integers and scaling them to the desired floating point values.

In the formulas below Vtrue denotes the true value of the data item, Vraw the one which is stored in the data element on file. The attributes are:

• transform: This is the indicator that a transformation of the Vraw data is necessary. Transform can have one the following values:
  • offset: Vtrue = Vraw + offset
  • scaling: Vtrue = Vraw * scaling
  • scaling_offset: both an offset and scaling is applied. Vtrue = Vraw*scaling + offset
  • sqrt_scaled: Vtrue = (Vraw/scaling)*(Vraw/scaling)
  • logarithmic_scaled: Vtrue = (Vraw/scaling)**10
  • polynomial: Vtrue = p1 + p2*Vraw + p3*Vraw*Vraw + p4*Vraw*Vraw*Vraw ….
• offset: The offset to apply
• scaling: The scale factor to apply
• direction: a komma separated list of length ndim which specifies for each dimension if it is increasing or decreasing. If this attribute is missing, increasing is implied.
• precedence: a komma separated list of length ndim which gives the rank order in which array indexes change with respect to other indexes. A precedence of 1 denotes the fastest changing index. If this attribute is missing, C storage order is implied.
• coefficients, a komma separated list of the polynomial coefficients to use for a polynomial transform

Update 01/2015

There was some discussion on this at NIAC 2010 and 2012. IMHO, the result was that all fixed schemes fall over in some point when people come up with new scaling schemes. It was decided to devise a NXformula base class to solve this problem. There was some further discussion on scaling in 2014 in the mailing list and on the teleconferences. At NIAC 2014 it was decided to accept a NXformula base class as suggested by Ben Watts as an experimental feature.