In the runs we used for the paper, the Gibbs reactor simulation converged in 570/625 samples. When I regenerate training data with IDAES 2.5.0, I get only 560 successful samples. The fewer samples cause the ALAMO surrogate to perform much worse. We should pin to specific Pyomo/IDAES versions and investigate the cause of this regression.
In the runs we used for the paper, the Gibbs reactor simulation converged in 570/625 samples. When I regenerate training data with IDAES 2.5.0, I get only 560 successful samples. The fewer samples cause the ALAMO surrogate to perform much worse. We should pin to specific Pyomo/IDAES versions and investigate the cause of this regression.