CLIVAR-PRP / ENSO_metrics

CMIP ENSO metrics python code
BSD 3-Clause "New" or "Revised" License
30 stars 15 forks source link

Make educational example of different normalisation, color bar, etc... #34

Closed eguil closed 5 years ago

eguil commented 5 years ago

Use a MC to show different options of portrait plots:

lee1043 commented 5 years ago

RMSE Portrait plot

Color for normalized metrics value

ENSO_perf_cmip_5and6_Rmse_PP_SidebySide_annotated

Color for NON-normalized metrics value

ENSO_perf_cmip_5and6_Rmse_PP_SidebySideColor-Not-Normalized_annotated

Non-RMSE Portrait plot

Color for normalized metrics value

ENSO_perf_cmip_5and6_nonRmse_PP_SidebySide_annotated

Color for NON-normalized metrics value

ENSO_perf_cmip_5and6_nonRmse_PP_SidebySideColor-Not-Normalized_annotated

lee1043 commented 5 years ago

@eguil @yyplanton @gleckler1 what do you think? Any feedback/suggestion welcome

gleckler1 commented 5 years ago

@eguil @yyplanton @gleckler1 Great examples Jiwoo. For the non-normalized RMSE you could make an effort to encapsulate more rows by change the color min/max to 0-5... with that more of the rows will probably not be blue but it will still clearly demonstrate the issue of comparing results from different metrics.

lee1043 commented 5 years ago

@gleckler1 great idea, which look like below.

non-normalized RMSE (example 2)

ENSO_perf_cmip_5and6_Rmse_PP_SidebySideColor-Not-Normalized_annotated

eguil commented 5 years ago
lee1043 commented 5 years ago

ENSO_perf_cmip_5and6_Rmse_PP_SidebySide_SidebySide_NormalizedByMaximum_annotated ENSO_perf_cmip_5and6_nonRmse_PP_SidebySide_SidebySide_NormalizedByMaximum_annotated

Another example: Normalized by maximum of each row. [E(model) - E(max)] / E(max)

eguil commented 5 years ago

why don't we get any reds ?

lee1043 commented 5 years ago

It is normalized as: [E(model) - E(max)] / E(max), which makes all numbers as negative. Please let me know if you were indicating in different way. I am not sure you meat [E(model) - E(median)] / E(max)?

gleckler1 commented 5 years ago

I think they meant E(model)/E(max) … no E(median). So some should be close to one.

lee1043 commented 5 years ago

ENSO_perf_cmip_5and6_Rmse_PP_SidebySide_SidebySide_NormalizedByMaximum_annotated ENSO_perf_cmip_5and6_nonRmse_PP_SidebySide_SidebySide_NormalizedByMaximum_annotated

Another example 2: Normalized by maximum of each row. E(model) / E(max)