- 2025/05/15: Finished v2 models. Trained using v1 as the initial model, and
performing kdescent fitting to the 4D space {log(M(t)), t, log(M_obs), t_obs}.
    - Model names:
        cenflow_v2_0.eqx
        satflow_v2_0.eqx
    - Respective creation scripts:
        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/kdescent_train.py cenflow_v2_0.eqx --num-kernels 20 --num-fourier-kernels 20 --plot-loss-curve --seed 2 --include-test --initial-model cenflow_v1_0-kde_20_20train.eqx --steps 1200 --learning-rate 3e-5

        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/kdescent_train.py satflow_v2_0.eqx --num-kernels 20 --num-fourier-kernels 20 --plot-loss-curve --seed 2 --include-test --initial-model satflow_v1_0-kde_20_20train.eqx --steps 1200 --learning-rate 3e-5 --sats


- 2025/04/27: Finished v1 models, using full training data. Trained using the
flowjax fitter only.
    - Model names:
        cenflow_v1_0.eqx
        satflow_v1_0.eqx
    - Respective creation scripts:
        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/flowjax_train.py cenflow_v1_0.eqx --max-epochs 200 --max-patience 25 --learning-rate 8e-5 --nn-depth 8 --nn-width 48 --seed 111 --include-test --initial-model cenflow_v1_0train.eqx

        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/flowjax_train.py satflow_v1_0.eqx --max-epochs 200 --max-patience 25 --learning-rate 8e-5 --nn-depth 8 --nn-width 48 --seed 111 --include-test --initial-model cenflow_v1_0train.eqx --sats

- 2025/04/21: Preliminary versions of v1 models, from the "train split" (~70%
of the full training data) only. Trained using the flowjax fitter only.
    - Model names:
        cenflow_v1_0train.eqx
        satflow_v1_0train.eqx
    - Respective creation scripts:
        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/flowjax_train.py cenflow_v1_0train.eqx --max-epochs 400 --max-patience 30 --learning-rate 8e-5 --nn-depth 8 --nn-width 48 --seed 110

        mpiexec -n 1 $watchmemory python -W ignore ~/local/diffmahnet/scripts/flowjax_train.py satflow_v1_0train.eqx --max-epochs 400 --max-patience 30 --learning-rate 8e-5 --nn-depth 8 --nn-width 48 --seed 110 --sats
