I am running PF2 on a large dataset (396 tips, 12K subsets). One thing that slows down the PF2 run is the branch length estimation step. In a test run (with a small rcluster-max), branch length estimation took 32 hours of the total 34.5 hours total PF2 run time. Branch length estimation on this dataset takes only < 1 hour in an HPC environment with MPI parallelization. Please consider making an enhancement to the code such that when branch lengths are supplied in the user_tree_topology portion of the config file (which in my opinion is best estimated with MPI parallelization for large datasets), the user supplied branch lengths are used and branch length estimation is skipped.
I am running PF2 on a large dataset (396 tips, 12K subsets). One thing that slows down the PF2 run is the branch length estimation step. In a test run (with a small rcluster-max), branch length estimation took 32 hours of the total 34.5 hours total PF2 run time. Branch length estimation on this dataset takes only < 1 hour in an HPC environment with MPI parallelization. Please consider making an enhancement to the code such that when branch lengths are supplied in the user_tree_topology portion of the config file (which in my opinion is best estimated with MPI parallelization for large datasets), the user supplied branch lengths are used and branch length estimation is skipped.