双变量CDF / CCDF分布Python

Question

I am trying to plot a bivariate ccdf of the dataset that has x and y values both.

Univariate I can plot very well, below is the input and the code is for univeriate dataset.

Input: These are only first 20 rows of the data points. Input has 1000s of rows and of which col[1] and col[3] needs to be plotted as they posses a user and keyword frequency relationship.

tweetcricscore  34 #afgvssco   51
tweetcricscore  23 #afgvszim   46
tweetcricscore  24 #banvsire   12
tweetcricscore  456 #banvsned  46
tweetcricscore  653 #canvsnk   1
tweetcricscore  789 #cricket   178
tweetcricscore  625 #engvswi   46
tweetcricscore  86 #hkvssco    23
tweetcricscore  3 #indvsban    1
tweetcricscore  87 #sausvsvic  8
tweetcricscore  98 #wt20       56

Code: univeriate dataset

import numpy as np
import matplotlib.pyplot as plt
from pylab import*
import math
from matplotlib.ticker import LogLocator

data = np.genfromtxt('keyword.csv', delimiter=',', comments=None)

d0=data[:,1]
X0 = np.sort(d0)
cdf0 = np.arange(len(X0))/float(len(X0))
ccdf0 = 1 - cdf0
plt.plot(X0,ccdf0, color='b', marker='.', label='Keywords')

plt.legend(loc='upper right')
plt.xlabel('Freq (x)')
plt.ylabel('ccdf(x)')
plt.gca().set_xscale("log")
#plt.gca().set_yscale("log")
plt.show()

I am looking for some option for bivariate data points. I referred Seaborn Bivariate Distribution But I am not able to put it in proper context with my dataset.

Any alternative suggestion within python, matplotlib, seaborn are welcome.. Thanks in advance.

Answer 1

Bivariate distributions the way you're trying to describe are oftentimes continuous, for instance the size of a house (input, x) and it's price (output, y.) In your case there is no meaningful relationship (I think) in the number of the keyword, as it's probably just an ID assigned to the keyword right?

In your case to me it seems as though you have categories (keywords). each category appears to have two numbers a tweetcricscore and a keyword number. \\

Your code here:

cdf0 = np.arange(len(X0))/float(len(X0))

To me suggests that your x range is just their labels and not a meaningful value.

A better source for categorical plots can be found here .

To create a bivariate distribution, assuming that's still what you want having read that, you'd do the following using your data as an example using your data from above:

import numpy as np
import seaborn as sns

col_1 = np.array([34, 23, 24, 456, 653, 789, 625, 86, 3, 87, 98])
col_3 = np.array([51, 46, 12, 46, 1, 178, 46, 23, 1, 8, 56])

sns.jointplot(x=col_3, y=col_1)

Which produces the very nonsensical figure here:

You'll have to add the x and y labels manually; this is because you're passing numpy array s instead of pandas Dataframes which can be thought of like dictionaries where each key in the dictionary is the title of a column, and the value the numpy array.

Using random numbers to show how it might look with a more random, continuous, related dataset.

This is the example taken from the docs.

import numpy as np
import seaborn as sns
import pandas as pd

mean, cov = [0, 1], [(1, .5), (.5, 1)]
data = np.random.multivariate_normal(mean, cov, 200)
df = pd.DataFrame(data, columns=["x", "y"])
sns.jointplot(x="x", y="y", data=df);

Which gives this:

The bar graphs on top of the chart can be thought of as uni variate charts (what you probably have produced) because they just describe the distribution of one or the other variable (x, or y, col_3, or col_1)