#!/usr/bin/env python
# coding: utf-8

# # Quick Play With EU Ref Data

# In[37]:


#data from http://www.electoralcommission.org.uk/find-information-by-subject/elections-and-referendums/upcoming-elections-and-referendums/eu-referendum/electorate-and-count-information


# In[38]:


get_ipython().system('ls EU*')


# Load the data in:

# In[39]:


import pandas as pd
#df=pd.read_csv('EU-referendum-result-data.csv')
url='http://www.electoralcommission.org.uk/__data/assets/file/0014/212135/EU-referendum-result-data.csv'
df=pd.read_csv(url)
df[:3]


# Check the columns:

# In[40]:


df.columns


# How close were the votes? Count number of areas where difference between remanin and leave was less than or equal to `2 * halfwidth` (i.e. both votes in band `50 +/- halfwidth`.

# In[51]:


halfwidth=1
df[(abs(df['Pct_Remain']-50)<halfwidth) & (abs(df['Pct_Leave']-50)<halfwidth)]


# Caluclate the spread(?) - difference between the remain and leave.

# In[42]:


df['spread']=df['Pct_Remain']-df['Pct_Leave']


# Plot the spread as a histogram, using quite tight bin widths.

# In[43]:


df['spread'].plot(kind='hist',bins=200, figsize=(20,10))


# How about a cumulative count of the absolute spread?

# In[44]:


abs(df['spread']).plot(kind='hist',bins=200, figsize=(20,10),cumulative=True)


# Do some nicer plots?

# In[45]:


import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt


# In[46]:


# Plot a historgram and kernel density estimate
sns.distplot(df['spread'], color="m",bins=100)


# Crib from the seaborn gallery to do a multiplot.

# In[47]:


f, axes = plt.subplots(2, 1, figsize=(7, 7), sharex=True)
sns.distplot(df['spread'], kde=False, color="b", ax=axes[0],bins=200)
sns.distplot(df['spread'], color="m", ax=axes[1],bins=200)


# How about the absolute spread?

# In[48]:


f, axes = plt.subplots(2, 1, figsize=(7, 7), sharex=True)
sns.distplot(abs(df['spread']), kde=False, color="b", ax=axes[0],bins=100)
sns.distplot(abs(df['spread']), color="m", ax=axes[1],bins=100)


# How about a cumulative kernel density estimate of the absolute spread?

# In[49]:


sns.kdeplot(abs(df['spread']), cumulative=True)


# What if we count over the actual (signed) spread? What percentage of reporting areas voted leave (negative)?

# In[50]:


sns.kdeplot(df['spread'], cumulative=True)


# In[ ]: