notebook.community

Edit and run 





In [ ]:



    


import numpy as np
import pandas as pd
import itertools
from __future__ import division
import geoplotlib as glp
import geoplotlib as gpl
from geoplotlib.utils import BoundingBox, DataAccessObject

%matplotlib inline
pd.set_option('display.max_columns', None)



    











In [ ]:



    






    











In [ ]:



    


filePath = 'datasets/NYPD_Motor_Vehicle_Collisions.csv'

collisions = pd.read_csv(filePath)
collisions = collisions[pd.notnull(collisions.BOROUGH)]
collisions.head()



    











In [ ]:



    






    











In [ ]:



    


collisions['CROSS STREET NAME'].value_counts()



    











In [ ]:



    


mask = ((collisions['CROSS STREET NAME'] == 'BUSHWICK AVENUE') & (collisions['ON STREET NAME'] == 'COOK STREET'))
#mask = ((collisions['CROSS STREET NAME'] == 'BUSHWICK AVENUE'))

bushwick = collisions[mask]
bushwick



    











In [ ]:



    


locationMask = (collisions.LOCATION == '(40.702713, -73.9373173)')

collisions[locationMask]



    











In [ ]:



    


locationsCounts = collisions.LOCATION.value_counts()
locationsCountsFrame  = locationsCounts.to_frame()
locationsCountsFrame
#locationsCountsFrame.count



    











In [ ]:



    


locationsCountsFrame['LAT'] = locationsCountsFrame.index.str.split(',').str.get(0).str.replace('(', '')
locationsCountsFrame['LON'] = locationsCountsFrame.index.str.split(',').str.get(1).str.replace(')', '')

#locationsCountsFrame.index



    











In [ ]:



    


#locationsCountsFrame

sums = collisions.groupby(['LOCATION'])[['NUMBER OF PERSONS INJURED', 'NUMBER OF PERSONS KILLED']].sum()



    











In [ ]:



    


result = locationsCountsFrame.combine_first(sums)
result



    











In [ ]:



    






    











In [ ]:



    


result.sort_values(by='LOCATION')



    











In [ ]:



    


intersections = result



    











In [ ]:



    


result.to_csv('datasets/collision_counts.csv', sep=',')



    











In [ ]:



    


collisions['year'] = collisions.DATE.str.split('/').str.get(2)

top10Intersections = collisions.LOCATION.value_counts()[:10]


top10 = collisions[collisions.LOCATION.isin(top10Intersections.index)]



    











In [ ]:



    






    











In [ ]:



    


top10Intersections.to_frame()



    











In [ ]:



    


def addStreetName(e):
    return top10[top10.LOCATION == e.name][['ON STREET NAME', 'CROSS STREET NAME']].iloc[0]
    
top10InDf = top10Intersections.to_frame()
    
    
top10InDf[['ON STREET NAME', 'CROSS STREET NAME']] = top10InDf.apply(addStreetName, axis=1)
top10InDf.rename(columns={'LOCATION': 'COLLISIONS'}, inplace=True)
top10InDf.index.name = "LOCATION"
top10InDf.to_csv('datasets/intersection_top_10.csv', sep=',')



    











In [ ]:



    


series = {};

for location in top10.LOCATION:
    a_location = top10[top10.LOCATION == location]
    #print a_location.year
    index = []
    data = []
    key = a_location['ON STREET NAME'].iloc[0] + ' / ' + a_location['CROSS STREET NAME'].iloc[0]
    for year in range(2012, 2016):
        index.append(year)
        data.append(len(a_location[a_location.year == str(year)].index))
    series[key] = pd.Series(data, index=index);

Here are the top 10 most dangerous intersection plotted by year

Note that 2012 and 2016 are not complete.





In [ ]:



    


pd.DataFrame(series).plot(kind='bar',figsize=(15,10), subplots=True, layout=(5,2), legend=False, sharey=True)

Here we see the location of these 10 intersection





In [ ]:



    


lat = top10.LATITUDE.values
lon = top10.LONGITUDE.values

glp.dot({'lat': lat, 'lon': lon}, color="r")
glp.inline()

lat

That the intersections with the most collions

However how many get injured or killed in these?





In [ ]:



    


top10Injuries = collisions.groupby(['LOCATION'])['NUMBER OF PERSONS INJURED'].sum().sort_values(ascending=False)[:10]
top10Injuries



    











In [ ]:



    


top10Injuries



    











In [ ]:



    


top10Killed = collisions.groupby(['LOCATION'])['NUMBER OF PERSONS KILLED'].sum().sort_values(ascending=False)[:11]
top10Killed



    











In [ ]:



    


def splitLocation(df):
    result = df.copy()
    result['LAT'] = result.index.str.split(',').str.get(0).str.replace('(', '')
    result['LON'] = result.index.str.split(',').str.get(1).str.replace(')', '')
    return result
    

top10KilledDf = splitLocation(top10Killed.to_frame())
top10InjuredDf = splitLocation(top10Injuries.to_frame())



    











In [ ]:



    


lon = top10KilledDf.LON.astype(float).values
lat = top10KilledDf.LAT.astype(float).values

glp.dot({'lat': lat, 'lon': lon}, color="r")
glp.dot({'lat': top10.LATITUDE.values, 'lon': top10.LONGITUDE.values}, color='g')
glp.dot({'lat': top10InjuredDf.LAT.astype(float).values, 'lon': top10InjuredDf.LON.astype(float).values}, color='b')

glp.inline()

Lets save them





In [ ]:



    


def get_intersection_name(e):
    name = collisions[collisions.LOCATION == e.name][['CROSS STREET NAME','ON STREET NAME']].iloc[0]
    return name['ON STREET NAME'] + ' / '+ name['CROSS STREET NAME']
    
top10InjuredDf['NAME'] = top10InjuredDf.apply(get_intersection_name, axis=1, raw=True)
top10KilledDf['NAME'] = top10KilledDf.apply(get_intersection_name, axis=1, raw=True)



    











In [ ]:



    


top10InjuredDf



    











In [ ]:



    


top10KilledDf



    











In [ ]:



    


top10InjuredDf.to_csv('datasets/intersection_top10_injuries.csv', sep=',')
top10KilledDf.to_csv('datasets/intersection_top10_killed.csv', sep=',')

Finding outliers

Finding the intersection with the most injured, relative to the collisions





In [ ]:



    


# Lets find the average collisions per intersection

means = intersection.mean()
means



    











In [ ]:



    


print "For each collision %f number of persons gets injured" % (means['NUMBER OF PERSONS INJURED'] / means.LOCATION)



    











In [ ]:



    


top10InjuredDf



    











In [ ]:



    


top10Means = top10.mean()
top10InjuredMeans = top10InjuredDf.mean()

top10KilledMean = top10KilledDf.mean()
print "In the intersections with the most collisions, only %f amount of persons where injured\n" % top10Means['NUMBER OF PERSONS INJURED']

print "However looking at the top most intersection with the most injured,\n then we have %f persons injured per collison.\n"\
% (top10InjuredMeans['NUMBER OF PERSONS INJURED'] / top10InjuredMeans['COLLISIONS'])

#print "And lastly looking at the intersections with the most killed people, only %f persons get injured (not including killed) per collision"\
#% (top10KilledMean['NUMBER OF PERSONS INJURED'] / top10KilledMean['COLLISIONS'])



    











In [ ]:



    


loc_count = collisions.LOCATION.value_counts()


def get_collision_count(e):
    return loc_count[e.name]
    

top10InjuredDf['COLLISIONS'] = top10InjuredDf.copy().apply(get_collision_count, axis=1)

top10KilledDf['COLLISIONS'] = top10KilledDf.copy().apply(get_collision_count, axis=1)



    











In [ ]:



    


collisions.groupby(['LOCATION']).value_counts()

Look for some intersections that have a perticular high ration

Meaning where is it perticular dangous to drive.





In [ ]:



    


def calc_collision_inju_ration(e):
    return (e['NUMBER OF PERSONS INJURED'] / e.LOCATION)

intersection[intersection.LOCATION > 20].apply(calc_collision_inju_ration, axis=1, raw=True).sort_values()



    











In [ ]:



    


collisions[collisions.LOCATION == '(40.6790346, -73.9385077)']



    











In [ ]:



    


top10['CONTRIBUTING FACTOR VEHICLE 1'].value_counts()



    











In [ ]:



    


def causesTop(n):
    danger_zone = collisions.LOCATION.value_counts()[n-1:n]
    return collisions[collisions.LOCATION.isin(danger_zone.index)]['CONTRIBUTING FACTOR VEHICLE 1'].value_counts()


series = {}

for i in range(1,10):
    series[i] = causesTop(i)
    
pd.DataFrame(series).plot(kind='bar', figsize=(15,10), subplots=True)



    











In [ ]:



    


numberTop6 = collisions[collisions.LOCATION.isin(collisions.LOCATION.value_counts()[5:6].index)]



    











In [ ]:



    


pd.to_datetime(numberTop6.DATE).value_counts().plot(figsize=(15,15))


#pd.Series(range(0,numberTop6['NUMBER OF PERSONS INJURED'].size), index=pd.to_datetime(numberTop6.DATE)).plot(kind='bar',figsize=(15,15))



    











In [ ]:



    


numberTop6.size



    











In [ ]:



    


top10



    











In [ ]:



    


#top10['HOUR'] = top10.TIME.str.split(':').str.get(0).astype('int')
#top1['HOUR'] = top1.TIME.str.split(':').str.get(0).astype('int')
collisions['HOUR'] = collisions.TIME.str.split(':').str.get(0).astype('int')



    











In [ ]:



    


top10.HOUR.value_counts().sort_index().plot(figsize=(15,10), kind='bar')



    











In [ ]:



    


top1.HOUR.value_counts().sort_index().plot(figsize=(15,10), kind='bar')



    











In [ ]:



    


top10.HOUR.value_counts().index



    











In [ ]:



    


top1 = collisions[collisions.LOCATION.isin(collisions.LOCATION.value_counts()[0:1].index)]



    











In [ ]:



    


series = {}

for r in collisions['CONTRIBUTING FACTOR VEHICLE 1'].unique():
    series[r] = collisions[collisions['CONTRIBUTING FACTOR VEHICLE 1'] == r].HOUR.value_counts().sort_index()
series['Combined'] = collisions.HOUR.value_counts().sort_index()
pd.DataFrame(series).plot(figsize=(15,100), subplots=True)



    











In [ ]:



    


collisions['CONTRIBUTING FACTOR VEHICLE 1'].unique()



    











In [ ]:



    


collisions



    











In [ ]:



    


collisions.['']

Content source: masve/saav-deliveries

Similar notebooks:

notebook.community | gallery | about