## 2D kernel density estimation and its Plotly plot¶

We have two Excel files with two columns. We read the files into two pandas dataframes and plot for each of them an estimate of the joint distribution of the corresponding two columns. The joint distribution is calcalutated by scipy.stats.gaussian_kde function.

In [23]:
import numpy as np
import pandas as pd
import seaborn as sns
import numpy as np
import scipy.stats as st

import matplotlib.pyplot as plt
%matplotlib inline


Read the first file:

In [24]:
xl = pd.ExcelFile("Data/CSCEng.xls")
dfc = xl.parse("Sheet1")
dfc.columns

Out[24]:
Index([u'multiannual', u'bachelor-th'], dtype='object')

and the seconed one:

In [25]:
xl = pd.ExcelFile("Data/SystEng.xls")
dfi = xl.parse("Sheet1")
dfi.columns

Out[25]:
Index([u'multiannual', u'bachelor-th'], dtype='object')

The contour plot of the joint distribution of two variables (columns) is colored with a custom colorscale:

In [26]:
cubehelix_cs=[[0.0, '#fcf9f7'],
[0.16666666666666666, '#edcfc9'],
[0.3333333333333333, '#daa2ac'],
[0.5, '#bc7897'],
[0.6666666666666666, '#925684'],
[0.8333333333333333, '#5f3868'],
[1.0, '#2d1e3e']]


The function kde_scipy returns data for Plotly contour plot of the estimated 2D distribution:

In [27]:
def kde_scipy( vals1, vals2, (a,b), (c,d), N ):

#vals1, vals2 are the values of two variables (columns)
#(a,b) interval for vals1; usually larger than (np.min(vals1), np.max(vals1))
#(c,d) -"-          vals2

x=np.linspace(a,b,N)
y=np.linspace(c,d,N)
X,Y=np.meshgrid(x,y)
positions = np.vstack([Y.ravel(), X.ravel()])

values = np.vstack([vals1, vals2])
kernel = st.gaussian_kde(values)
Z = np.reshape(kernel(positions).T, X.shape)

return [x, y, Z]


### Contour plot of the joint distribution of data from the first file¶

In [28]:
import plotly.plotly as py
from plotly.graph_objs import *

In [29]:
def make_kdeplot(varX, varY, (a,b), (c,d), N, colorsc, title):
#varX, varY are lists, 1d numpy.array(s), or dataframe columns, storing the values of two variables

x, y, Z = kde_scipy(varY, varX, (a,b), (c,d), N )

data = Data([
Contour(
z=Z,
x=x,
y=y,
colorscale=colorsc,
#reversescale=True,
opacity=0.9,
contours=Contours(
showlines=False)
),
])

layout = Layout(
title= title,
font= Font(family='Georgia, serif',  color='#635F5D'),
showlegend=False,
autosize=False,
width=650,
height=650,
xaxis=XAxis(
range=[a,b],
showgrid=False,
nticks=7
),
yaxis=YAxis(
range=[c,d],
showgrid=False,
nticks=7
),
margin=Margin(
l=40,
r=40,
b=85,
t=100,
),
)

return Figure( data=data, layout=layout )

In [30]:
N=200
a,b=(5,11)
fig=make_kdeplot(dfc['multiannual'], dfc['bachelor-th'], (a,b), (a,b),
N, cubehelix_cs,'kde plot of two sets of data' )

py.sign_in('empet', 'my_api_key')
py.iplot(fig, filename='kde-2D-CSCE')

Out[30]:

### Contour plot of the joint distribution of data from the second file¶

In [31]:
a, b=(4,12)
fig=make_kdeplot(dfi['multiannual'], dfi['bachelor-th'], (a,b), (a,b),
N, cubehelix_cs, 'kde plot of two sets of data')
py.iplot(fig, filename='kde-2D-SE')

Out[31]:

One notices that the second contourplot illustrates a mixture of two bivariate distributions.

Finally we read a dataframe from a csv file posted on the Plotly's github account, select the rows corresponding to Iris-virginica, and plot the joint distribution of two virginica features:

In [32]:
df = pd.read_csv('https://raw.githubusercontent.com/plotly/datasets/master/iris.csv')
virginica = df.loc[df.Name == "Iris-virginica"]
a, b=(5,8.5)
c,d=(2,4)
N=100
fig=make_kdeplot(virginica.SepalLength, virginica.SepalWidth, (a,b), (c,d),
N, cubehelix_cs, 'kde plot of joint distribution for virginica SepalLength and SepalWidth')
py.iplot(fig,  filename='virginica-sepal-length-vs-width')

Out[32]:
In [33]:
from IPython.core.display import HTML
def  css_styling():
styles = open("./custom.css", "r").read()
return HTML(styles)
css_styling()

Out[33]: