Plotting Multiple Histograms in Matplotlib - Colors or side-by-side bars

Question

Problem : When Plotting Multiple Histograms in Matplotlib, i cannot differentiate a plot from another

Problem as Image : ** **Minor Problem : The left label 'Count' is out of the image, partially. Why?

Description

I want to plot the histogram of the 3 different sets. Each set, is an array with 0's and 1's. I want the histogram of each so i can detect imbalances on the dataset.

I have them plotted separately but i wanted a graphic of them together.

It would be okay to have a different graphic with bars side-by-side or, i even googled about plotting it as 3D, but i dont know how easy would be to "read" or "look" at the graphic and understand it.

Right now, i want to plot the [train],[validation] and [test] bars at each side on the same graphic, something like this :

PS : My googling didnt return any code that was understandable to me. Also, i would like if someone would check if im doing any insanity on my code.

Thanks a lot guys!

Code :

def generate_histogram_from_array_of_labels(Y=[], labels=[], xLabel="Class/Label", yLabel="Count", title="Histogram of Trainset"):
    plt.figure()
    plt.clf()

    colors = ["b", "r", "m", "w", "k", "g", "c", "y"]

    information = []
    for index in xrange(0, len(Y)):
        y = Y[index]

        if index > len(colors):
            color = colors[0]
        else:
            color = colors[index]

        if labels is None:
            label = "?"
        else:
            if index < len(labels):
                label = labels[index]
            else:
                label = "?"

        unique, counts = np.unique(y, return_counts=True)
        unique_count = np.empty(shape=(unique.shape[0], 2), dtype=np.uint32)

        for x in xrange(0, unique.shape[0]):
            unique_count[x, 0] = unique[x]
            unique_count[x, 1] = counts[x]

        information.append(unique_count)

        # the histogram of the data
        n, bins, patches = plt.hist(y, unique.shape[0], normed=False, facecolor=color, alpha=0.75, range=[np.min(unique), np.max(unique) + 1], label=label)

    xticks_pos = [0.5 * patch.get_width() + patch.get_xy()[0] for patch in patches]

    plt.xticks(xticks_pos, unique)

    plt.xlabel(xLabel)
    plt.ylabel(yLabel)
    plt.title(title)
    plt.grid(True)
    plt.legend()
    # plt.show()

    string_of_graphic_image = cStringIO.StringIO()

    plt.savefig(string_of_graphic_image, format='png')
    string_of_graphic_image.seek(0)

    return base64.b64encode(string_of_graphic_image.read()), information

Edit

Following the answer of hashcode, this new code :

def generate_histogram_from_array_of_labels(Y=[], labels=[], xLabel="Class/Label", yLabel="Count", title="Histogram of Trainset"):
    plt.figure()
    plt.clf()

    colors = ["b", "r", "m", "w", "k", "g", "c", "y"]
    to_use_colors = []
    information = []


    for index in xrange(0, len(Y)):
        y = Y[index]

        if index > len(colors):
            to_use_colors.append(colors[0])
        else:
            to_use_colors.append(colors[index])

        unique, counts = np.unique(y, return_counts=True)
        unique_count = np.empty(shape=(unique.shape[0], 2), dtype=np.uint32)

        for x in xrange(0, unique.shape[0]):
            unique_count[x, 0] = unique[x]
            unique_count[x, 1] = counts[x]

        information.append(unique_count)

    unique, counts = np.unique(Y[0], return_counts=True)
    histrange = [np.min(unique), np.max(unique) + 1]
    # the histogram of the data
    n, bins, patches = plt.hist(Y, 1000, normed=False, alpha=0.75, range=histrange, label=labels)


    #xticks_pos = [0.5 * patch.get_width() + patch.get_xy()[0] for patch in patches]

    #plt.xticks(xticks_pos, unique)

    plt.xlabel(xLabel)
    plt.ylabel(yLabel)
    plt.title(title)
    plt.grid(True)
    plt.legend()

Is producing this :

-- New Edit :

def generate_histogram_from_array_of_labels(Y=[], labels=[], xLabel="Class/Label", yLabel="Count", title="Histogram of Trainset"):
    plt.figure()
    plt.clf()

    information = []

    for index in xrange(0, len(Y)):
        y = Y[index]

        unique, counts = np.unique(y, return_counts=True)
        unique_count = np.empty(shape=(unique.shape[0], 2), dtype=np.uint32)

        for x in xrange(0, unique.shape[0]):
            unique_count[x, 0] = unique[x]
            unique_count[x, 1] = counts[x]

        information.append(unique_count)

    n, bins, patches = plt.hist(Y, normed=False, alpha=0.75, label=labels)

    plt.xticks((0.25, 0.75), (0, 1))

    plt.xlabel(xLabel)
    plt.ylabel(yLabel)
    plt.title(title)
    plt.grid(True)
    plt.legend()

Is working now but, the label from the left side is kinda out of bounds and i wanted to center the bars better... How can i do that?

Result :

Tinney

You have removed the bins param, by default it is set to 10. Just add a bins param like this - n, bins, patches = plt.hist(Y, bins = 2, normed=False, alpha=0.75, range=histrange, label=labels) – Strappado

Have you tried setting the bins to 2? – Strappado

And regarding that label not being displayed, I guess its a machine specific problem. You can try adjusting the subplot...Check this out matplotlib.org/examples/pylab_examples/subplots_adjust.html – Strappado

it worked! Hm just one more question if you dont mind.. can i make the graphic bigger? Like the image width *2 and height * 2 – Tinney

Yeah you can do it by setting the rcparams... from matplotlib import rcParams – Strappado

rcParams['figure.figsize'] = (20, 20) adjust the value accordingly – Strappado

The default is rcParams['figure.figsize'] = (1, 1) ? – Tinney

(20, 20) is the size in inches, (1, 1) will produce a tiny plot :P. I don't know the default, but it might be (5, 5). – Strappado

Answer 1

I tried and came up with this. You can change the xticks position in the code. Simply what you have to do is pass on a tuple to the plt.hist, can't be more simple right !? So lets suppose you have two lists of 0s and 1s, so what you gotta do is -

a = np.random.randint(2, size=1000)
b = np.random.randint(2, size=1000)
plt.hist((a, b), 2, label = ("data1", "data2"))
plt.legend()
plt.xticks((0.25, 0.75), (0, 1))

The exact code I tried to run (after changing the number of bins to 2)-

a = np.random.randint(2, size=1000)
b = np.random.randint(2, size=1000)
y = [a, b]
labels = ["data1", "data2"]
generate_histogram_from_array_of_labels(Y = y, labels = labels)

Aand I got the same result...

Strappado

It seems it did work now! However, can you help me fix this minor issues? I wanted to center the bars better on the xlabels! Also the left label is out of bounds! – Tinney

Answer 2

If your datasets are of equal length, you might be able to do this easily with pandas. So assuming you have

import numpy

N = 1000
train, validation, test = [numpy.random.randint(2, size=N) for _ in range(3)]
Y = [train, validation, test]

You can simply do

import pandas

df = pandas.DataFrame(list(zip(*Y)), columns=['Train', 'Validation', 'Test'])
df.apply(pandas.value_counts).plot.bar()

which results in this plot:

If you also import seaborn, it looks a bit nicer:

Spawn

I need to save the image.. is it possible with this code you presented? – Tinney

@ScientistGirl yes, just use savefig as usual. – Spawn