Reproducing published results¶

The scripts below live in bash_scripts/ of the repository and run from its root.

OpenDPDv1¶

To reproduce the PA modeling results shown in OpenDPD Figure 4(a):

bash bash_scripts/train_all_pa.sh

This script trains multiple PA models (each with approximately 500 parameters) using 5 different random seeds. Figure 4(a) displays the averaged results from these runs.

To reproduce the DPD learning results in Figure 4(b), Figure 4(d), and Table 1:

bash bash_scripts/train_all_dpd.sh

This script trains various DPD models, each with approximately 500 parameters.

Mixed-Precision DPD (MP-DPD)¶

For convenience, you can reproduce all MP-DPD results using:

bash bash_scripts/quant_mp_dpd.sh

OpenDPDv2¶

To reproduce the quantized (W16A16) TRes-DeltaGRU-450 DPD modeling results shown in OpenDPDv2 Table 1:

bash bash_scripts/OpenDPDv2.sh