Support for estimation output file with 32-bit floating point numbers

[hdante]

Hello, does CatEstimator (and child classes) support generating the estimate file using 32-bit numbers instead of 64-bit ? I believe that will be necessary at the LineA lab to reduce the output file sizes. The extra significant digits probably will not be necessary.

[eacharles]

We do have lossy compressed in qp. We could add some sort of compression flag to hdf5, but I'm not sure it would do all that much, as compression floats doesn't always get you much.

[hdante]

Hello, Eric, the instantaneous 50% size reduction when storing the output with 32-bit floating point numbers is guaranteed to be beneficial to the LineA lab. For example, large (10s of billions) estimations with interpolated results would reduce in size from, for example, almost 90 TB to 45 TB.

The benefit for format-specific compression flags is unknown because its biggest benefit requires the truncation provided by the 32-bit floating point numbers.

Note: I'm referring to lossless encoding only.