File exchange format

Yael uses a simple file format to store byte, integer or floating point vectors. Each vector is stored by

• writing its length d as an integer (4 bytes, little-endian)
• writing its d components in binary format.

Until now, we have used only single-precision floating point numbers. Therefore the size of storing an integer or a floating-point vector is simply sizeof(int) + 4 * d :math:` = 4 (d+1)`, as both the dimension and the components require 4 bytes each. For byte vectors, it is similarly equal to 4+d.

By convention we will call a file containing several vectors of the same kind ‘fvecfile’, ‘ivecfile’ and ‘bvecfile’.

Matrices are stored as a concatenation of vectors, again without a header. A float matrix is therefore a fvecfile where all vectors are the same length. Its suggested filename extention is .fvecs.

Some particular high-level functions are defined for these files.

Remarks:

• Since there use no header, several files of same type can be concatenated using regular the cat Unix command to produce a file containing the concatenation of the vector.

• in the progs subdirectory, the utilities xvecfile (x= f, i or b) receive a xvecfile on standard input and produce a text output. Usage:

  cat myfile.fvecs | fvecfile

  or, equivalently:

  fvecfile < myfile.fvecs
  

• The Yael functions used to read these the files in C, Python and Matlab are called Xvecs_read (x= f, i or b). The prototype of the function depends on whether you call it from C, Python or Matlab. See the I/O interface in vector.h.