codec / near-lossless / lossy
quant_linear
Allow the same absolute error everywhere.
quant_linear snaps values to evenly spaced levels around zero. If
MAX_ERROR=2, every reconstructed value stays within two units of the original, whether
that value is 10 or 10,000. Use it when a fixed error in the data's own units is easy to explain
and accept.
- ctid
- 0x72D781
- header
- dtype + flags + step · 10 bytes
- recipe
- LINEAR:MAX_ERROR=V · ,STORE=VALUES
- types
- integers 8 to 64-bit · floats 16 to 64-bit
- decode
- x = q · step, or a copy when the stream holds the reconstruction
Code
error="LINEAR:MAX_ERROR=5"
