Fig. 4. Average (±1SD) prediction accuracies for all participants and held-out conditions at a prediction horizon of 12.5% of a stride. Gray dots represent individual legs. Colored brackets denote statistically significant differences between held-out conditions for each model. Black horizontal bars denote significant differences between models across all three (solid) or two (dashed) held-out conditions. The large variance in the LPV model’s predictions of rectus femoris and gastrocnemius responses in the held-out K3 condition were due to bad predictions (RRV > 2) in a small number of legs. All comparisons used paired t-tests (α = 0.05) with a Holm-Sidak step-down correction for multiple comparisons.
...read more