A IPython Notebook to analyze the Gaza-Israel 2012 crisis

The Guardian is tracking and mapping live (link) the recent incidents in Gaza and Israel. As part of their data-journalism spirit, they are sharing the data as a Google Fusion Table available for access.

This notebook is an attempt to show, on the one hand, how the toolkit from the Python stack can be used for a real world data hack and, on the other, to offer deeper analysis beyond mapping of the events, both exploiting the spatial as well as the temporal dimension of the data.

  • The source document (.ipynb file) is stored on Github as a gist here, which means you can fork it and use it as a start for you own data-hack.
  • A viewable version is available here, via the IPython Notebook Viewer.

Collaborate on the notebook!!!

In its initial version (Nov. 20th), the notebook only contains code to stream the data from the Google Fusion Table into a pandas DataFrame (which means you get the data ready to hack!). Step in and collaborate in making it a good example of how Python can help analyze real world data. Add a new view, quick visualization, summary statistic of fancy model that helps understand the data better!

To contribute, just fork the gist as you would with any git repository.

Happy hacking!

In [18]:
%matplotlib inline
import matplotlib.pyplot as plt
import datetime
import urllib2, urllib
import pandas as pd
from StringIO import StringIO

The following cell pulls the data using the API. In the meantime, Google has changed its terms and ways to access it, so this might not work.

In []:
# Trick from http://stackoverflow.com/questions/7800213/can-i-use-pythons-csv-reader-with-google-fusion-tables

request_url = 'https://www.googleapis.com/fusiontables/v1/query'
query = 'SELECT * FROM 1KlX4PFF81wlx_TJ4zGudN_NoV_gq_GwrxuVau_M'

url = "%s?%s" % (request_url, urllib.urlencode({'sql': query}))
serv_req = urllib2.Request(url=url)
serv_resp = urllib2.urlopen(serv_req)
table = serv_resp.read()
print '\nLast pull of data from the Google FusionTable: ', datetime.datetime.now()
In [14]:
def parse_loc(loc, ret_lon=True):
    try:
        lon, lat = loc.split(',')
        lon, lat = lon.strip(' '), lat.strip(' ')
        lon, lat = map(float, [lon, lat])
        if ret_lon:
            return lon
        else:
            return lat
    except:
        return None
In []:
db = pd.read_csv(StringIO(table))

If you cannot pull the data using the API, an easy alternative is to export the table to a csv file manually and read it separately:

In [12]:
db = pd.read_csv('/Users/dani/Desktop/Gaza and Israel incidents mapped.csv')
In [16]:
db['lon'] = db['Location (approximate)'].apply(lambda x: parse_loc(x))
db['lat'] = db['Location (approximate)'].apply(lambda x: parse_loc(x, ret_lon=False))
db['Date'] = db['Date'].apply(pd.to_datetime)
db.head()
Out[16]:
Date Day Name of place Location (approximate) Details Source url Image url Icon 1 lon lat
0 2012-11-16 Friday Beit Lahia 31.5515, 34.5089 Firefighters try to extinguish a fire at a fac... http://www.guardian.co.uk/news/2012/nov/16/pic... http://static.guim.co.uk/sys-images/Guardian/P... placemark_circle_highlight 31.551500 34.508900
1 2012-11-15 Thursday Police Station in Deir al-Balah 31.4205, 34.3513 Israeli aircraft also bombed a police station ... Wires NaN placemark_circle_highlight 31.420500 34.351300
2 2012-11-15 Thursday Beit Hanoun 31.5382, 34.5380 Brothers Tareq Jamal Naser, 16, and Oday Jamal... http://www.maannews.net/eng/ViewDetails.aspx?I... NaN placemark_circle_highlight 31.538200 34.538000
3 2012-11-15 Thursday Sheikh Radwan neighborhood 31.536297, 34.465828 Violent explosions across Gaza City's Sheikh R... http://www.maannews.net/eng/ViewDetails.aspx?I... NaN placemark_circle_highlight 31.536297 34.465828
4 2012-11-15 Thursday Tel Aviv 32.0718, 34.777 Two rockets from Gaza crashed near Tel Aviv on... http://www.guardian.co.uk/world/2012/nov/15/is... NaN placemark_circle_highlight 32.071800 34.777000

Very basic descriptive analysis

  • Volume of incidents by day
In [20]:
t = db['Date']
t = t.reindex(t)
by_day = t.groupby(lambda x: x.day).size()
by_day.plot(kind='bar')
plt.title('Number of events by day')
plt.show()
  • Location of events coloured by day
In [29]:
f = plt.figure(figsize=(10, 6))
ax = f.add_subplot(111)
x, y = db['lon'], db['lat']
s = plt.scatter(x, y, marker='.', color='k')
for d, day in db.set_index('Date').groupby(lambda x: x.day):
    x, y = day['lon'], day['lat']
    c = cm.Set1(d/30.)
    s = plt.scatter(x, y, marker='^', color=c, label=str(d), s=20)
ax.get_yaxis().set_visible(False)
ax.get_xaxis().set_visible(False)
plt.legend(loc=2)
plt.title('Spatial distribution of events by day')
ax.set_axis_bgcolor("0.2") 
In [28]:
# You'll need cartopy for a pretty map
import cartopy.crs as ccrs
import cartopy.io.img_tiles as cimgt
import matplotlib.cm as cm
In [49]:
bg = cimgt.OSM()
src = ccrs.PlateCarree()

f = plt.figure(figsize=(20, 30))
ax = plt.axes(projection=bg.crs)
ax.add_image(bg, 9, alpha=0.5)

x, y = db['lon'], db['lat']
extent = [y.min(), y.max(), x.min(), 34]
extent = [34, 36, x.min(), x.max()] #Manually tweaked
for d, day in db.set_index('Date').groupby(lambda x: x.day):
    y, x = day['lon'], day['lat']
    c = cm.Set1(d/30.)
    s = plt.scatter(x, y, marker='^', color=c, label=str(d), s=40, \
                    transform=src)
ax.set_extent(extent, crs=src)
plt.legend(loc=2)
plt.title('Spatial distribution of events by day')
plt.show()