I’m working on a temporal prediction model predicting the state of a graph for several timesteps. I calculate all metrics per prediction distance. That is, for predicting the first, second, third… timestep after the input graph. I’d like to plot the errors with this prediction distance as x-axis. However from what I found I can only use step, batch or epoch as horizontal axis. As I calculate all the metrics in the test step, step, batch and epoch are the same so the scalars are all on one vertical line. Is there a way to provide a custom x-value to the tensorboard logger?