Matplotlib's legend: how to order entries by row first, rather than by column

Question

By default Matplotlib orders the entries in the legend by column, e.g.,

In [48]: from matplotlib.pyplot import subplots, show
    ...: f, ax = subplots(figsize=(6, 2), constrained_layout=True)
    ...: for i in range(1, 10):
    ...:     ax.plot((0,1),(0,1), label=str(i))
    ...: ax.legend(ncol=4, loc=3)
    ...: show()

but I'd like better to have the entries listed by rows, like

1  2  3  4
5  6  7  8
9

I could reorder the handles and labels and invoke ax.legend passing the reordered lists (I'll post that as an answer if requested in comments) but I'd like to know if there is a more direct and straightforward way to instruct Matplotlib to do as I want, so that I don't have to copy and paste the reorder solution in 50% of the Matplotlib scripts I'll write…

https://stackoverflow.com/a/22276133/6361531 I am afraid you might have to manually order your legend handlers. — Scott Boston, Mar 24 '21 at 14:34
@ScottBoston directly from the mouth of T.A. Caswell, ok, ok! I'll do it manually… — gboffi, Mar 24 '21 at 14:39
That is an old post 2014. I was looking in the recent version of the API documentation and I haven't found any help yet. — Scott Boston, Mar 24 '21 at 14:42
Even a reddit post 2 years ago https://www.reddit.com/r/learnpython/comments/8zbbql/changing_order_of_items_in_legend_matplotlibpyplot/ — Scott Boston, Mar 24 '21 at 14:44

gboffi · Accepted Answer · 2023-06-22T10:18:44.070

You can get the lists of handles to artists and the labels using the Axes method get_legend_handles_labels, then you call legend passing the reordered lists.

To reorder the lists, we observe that

legend places in the first column the first elements in the lists, then proceeds with the following columns/items,
the columns have variable lengths but all the rows must have the same length (i.e., ncol) except the last one, moreover in each column the increment betweeen items must equal ncol.

That said, it is possible to generate a sublist for each column using slicing, the length of the column is automatically the right one..., and eventually use sum to join the sub lists in the reordered one.

In [59]: from matplotlib.pyplot import subplots, show
    ...: reorder = lambda l, nc: sum((l[i::nc] for i in range(nc)), [])
    ...: 
    ...: f, ax = subplots(figsize=(6, 2), constrained_layout=True)
    ...: for i in range(1, 10):
    ...:     ax.plot((0,1),(0,1), label=str(i))
    ...: h, l = ax.get_legend_handles_labels()
    ...: ax.legend(reorder(h, 4), reorder(l, 4), ncol=4, loc=3)
    ...: show()

A few years later...

Slightly more compact in its usage when calling legend

reorder=lambda hl,nc:(sum((lis[i::nc]for i in range(nc)),[])for lis in hl)
...
h_l = ax.get_legend_handles_labels()
ax.legend(*reorder(h_l, 4), ncol=4)

(but you must remember to unpack the values returned by reorder).

Matplotlib's legend: how to order entries by row first, rather than by column

1 Answers1

Linked

Related