Hi,
I wanted to share that I recently updated the GPI / PI module in my package, Skyborn, to stay closely aligned with tcpyPI.
Skyborn now exposes decomposition-related functionality corresponding to:
log_decompose_pi
pi_log_decomposition
These were added to match the newer tcpyPI workflow more closely.
Consistency with tcpyPI
I validated the Skyborn results directly against tcpyPI outputs using the sample/reference data from tcpyPI.
Current comparison shows that the agreement is very tight:
- core PI diagnostics remain highly consistent with
tcpyPI
- RMSE for the main PI outputs is now on the order of
1e-4
- log-decomposition diagnostics also match very closely
Performance
Because Skyborn uses a compiled Fortran backend with whole-field processing, the same workflow is currently about 25x faster than point-by-point evaluation against tcpyPI in my benchmark setup.

Hi,
I wanted to share that I recently updated the GPI / PI module in my package, Skyborn, to stay closely aligned with
tcpyPI.Skyborn now exposes decomposition-related functionality corresponding to:
log_decompose_pipi_log_decompositionThese were added to match the newer
tcpyPIworkflow more closely.Consistency with tcpyPI
I validated the Skyborn results directly against
tcpyPIoutputs using the sample/reference data fromtcpyPI.Current comparison shows that the agreement is very tight:
tcpyPI1e-4Performance
Because Skyborn uses a compiled Fortran backend with whole-field processing, the same workflow is currently about 25x faster than point-by-point evaluation against
tcpyPIin my benchmark setup.