antoinecarme/pyaf

Forecast Quantiles Plots can be improved

antoinecarme opened this issue · 6 comments

antoinecarme commented

By looking at this jupyter notebook (large horizon models), in some forecast quantiles plots, the histograms are not visible at all

https://github.com/antoinecarme/pyaf/blob/3b6499491fd1ea2b237e5dae316591dfbde6af45/notebooks_sandbox/large_horizon/long_term_yosemite_tests.ipynb

image

antoinecarme commented

Proposed improvements :

  1. Better separate histograms (original issue solution). Assign a fixed height to each histogram.
  2. Use a more meaningful color map (gradient, Blue = Low, Green = Normal, Red = High) for synchronized histograms.
  3. Remove some unnecessary borders and yticks.
antoinecarme commented

Previous forecast quantiles plot (Ozone dataset)

image

New forecast quantiles plot (Ozone dataset)

image

antoinecarme commented

New forecast quantile plot for large horizon model (H = 512) from the jupyter notebook

image

antoinecarme commented

Further investigation shows that it is better to use the pre-defined matplotlib color map named 'turbo'.

image

This color map is the closest to the requested behavior and which has a more continuous color structure,

image

antoinecarme commented

CLOSED.

antoinecarme commented

Add a final touch for reducing unnecessary colors. Perform a quantization on 'turbo' color map.

image

image