matplotlib 'bar' vs 'barh' plots. Problem with barh

Question

I've just noticed something strange and was wondering if someone could explain what's going on?

I have a dictionary i'm plotting a horizontal and vertical bar charts from.

plt.bar(food.keys(), food.values()) #works fine, but:
plt.barh(food.keys(), food.values() #gives "unhashable type: 'dict_keys'" error.

if dictionary is unhashable, why does it let me plot a normal bar graph? Is this just a quirk of the barh function or am I doing something wrong?

Here's my test dataset:

food = {'blueberries':2, 'pizza':3.50, 'apples':0.50}

Thanks.

Answer 1

The other answer is of course correct in how to solve the problem - simply convert to a list. The reason why it doesn't work in the first place involves digging into the matplotlib source code.

bar and barh do both call matplotlib.axes.Axes.bar under the hood. When you plot a bar, x is the x position of the bars and height is the y value.

However, calling a barh, the y values (i.e. ['blueberries', 'pizza', 'apples']) is actually set to the argument bottom and the length of the bars is given by width.

The following line is called within the source code

func(ax, *map(sanitize_sequence, args), **kwargs)

Where sanitize_sequence is:

def sanitize_sequence(data):
    """
    Convert dictview objects to list. Other inputs are returned unchanged.
    """
    return (list(data) if isinstance(data, collections.abc.MappingView)
            else data)

The problem is that the argument bottom is a kwarg, and is therefore not passed to sanitize_sequence and is never converted to a list. Whereas in a normal bar, x is converted to a list by sanitize_sequence

So the issue is with how barh is implemented under the hood.

Answer 2

Nice MWE. Looks like you ran into a subtle python 2 to python 3 conversion bug.

Under python 2, dict.keys() (etc.) used to return lists of values. Under python 3, dict.keys() returns a view of the data corresponding to the keys. This view is not hashable (I think in principle, it would be; however, that behaviour is not implemented (yet)). Coercing your key view to a list circumvents the issue:

plt.barh(list(food.keys()), food.values())

Edit

The real question is why plt.bar works, as both, plt.bar and plt.barh call matplotlib.axes.Axes.bar under the hood. The definition of barh is literally:

def barh(self, y, width, height=0.8, left=None, *, align="center",
         **kwargs):
    kwargs.setdefault('orientation', 'horizontal')
    patches = self.bar(x=left, height=height, width=width, bottom=y,
                       align=align, **kwargs)
    return patches