Order bar charts

[Akaban]

Hello,

I would like to reopen issue 94 I don't think it's really user friendly to not include automatic sorting in geom_bar function. If I understand correctly we need to make a pandas.Categorical each time we want to define the order of our graph.

That's something I find really painful. Can't we implement a keyword_argument sort_by="xxxx" in geom_bar?

I can do the MR if it's not too much complicated, I don't know plotnine code as of now :)

Thanks for the answer

By the way thanks for this incredible package, easily the best python package for graphs available :)

[Akaban]

I think we should at least respect the order of the dataframe, it does not seem to be the case today

EDIT: Okay actually data is sorted before ggplot rendering, but it sort the x-axis alphabetically. We then should be able to pass sort=False in some kind of way to respect the order of the dataframe.

[Akaban]

For readers just so you know I found a "hack" to get cool bar plots without much effort.

1/ Sort your dataframe as you wish
2/ Compute a column rank which indicates the rank of the line (you may want to do this with group by)
3/ Compute new string column df["column"] = df["rank"].map(chr) + " - " + df["column"]

This ensure the alphabetic ordering of the column matches the order of rank thanks to the chr function, if add an extra useless character in rendering though.

:) I would still have preferred an explicit way to do this through plot nine

[TyberiusPrime]

Plotnine has inherited this from ggplot2, which respects the sort order of the datatypes involved, not the order of the data points.
I actually think that's a perfectly valid point of view that leads to fewer unexpected changes in plots.

Turning a column in a sorted dataframe into a categorical isn't that bad, though a bit verbose:
df.assign(column = pd.Categorical(df.column, df.column)).

Guess I'll be adding a convenient method for that in my dppd library...

[has2k1]

Sorting is standard and is handled automatically by the scale depending on the datatype (mentioned by @TyberiusPrime) , not the geom. Plus the scales are independent, the x scale knows nothing about any other column!

[isabelizimm]

For those stumbling upon this issue recently, you can do something like

(data
 >> ggplot()
        + geom_col(aes(x = 'reorder(col1, col2)', y = 'col2'))
)

after you have reordered your data as you wish.