What is the x-axis here? I trained for 100 epochs on a dataset of 1201 rows with default 70:20:10 split
the x increment is 1015, but I can’t connect that to any parameters… Batch size was 8… so 0.7 *1201/8 = 105 rows per training epoch… is that close enough?
PS I think the Max and Average labels are swapped…
Hey @JulianSMoore,
The increments themselves have no real meaning here, it’s just some intervals that are visually displayed, but you still have every other iteration between it.
Be mindful also that the gradients include both training and validation (something we do need to change) meaning that it would cover 135 rows per training epoch.
I’m a little surprised why the last value is 13485 rather than 13500 though.
For the other issue: it certainly looks like the Max and Average is swapped, although for some reason that’s not reproduced in all cases 
Added it as a bug we will look into, thanks for highlighting it! 
OK, let me think about those specific numbers - I’ll come back if it turns out to be important (I doubt it 
)
Strange that the Max/Av swap does not occur in all situations!