## Introduction
​
This notebook is the result of a short investigation of the 2015 Pronto Data Challenge.
It is intended to be a demonstration of how IPython/Jupyter notebooks
can be easily combined with the pandas, spark and d3 libraries based on previously existing demonstrations.
​
This notebook contains queries to generate two charts.
One explores trip flows between different regions of the Pronto system.
The second recreates a chart for a performance metric that has been measured in other bike share systems in the US.
​
Hopefully this notebook can be used to gain some quick insights into Pronto's first year, but it it is intended more to be a starting point for more queries.

Introduction

This notebook is the result of a short investigation of the 2015 Pronto Data Challenge. It is intended to be a demonstration of how IPython/Jupyter notebooks can be easily combined with the pandas, spark and d3 libraries based on previously existing demonstrations.

This notebook contains queries to generate two charts. One explores trip flows between different regions of the Pronto system. The second recreates a chart for a performance metric that has been measured in other bike share systems in the US.

Hopefully this notebook can be used to gain some quick insights into Pronto's first year, but it it is intended more to be a starting point for more queries.

 
## Library Initialization
​
To begin with, the system needs to be initialized and the libraries loaded.  This notebook relies on a combination of Pandas dataframes and Spark SQL queries to generate data for the D3 javascript and Matplotlib libraries.  It is based on examples that are referenced in the code comments.
​
This section can be skipped if you are more interested in the results than the underlying technology.

Library Initialization

To begin with, the system needs to be initialized and the libraries loaded. This notebook relies on a combination of Pandas dataframes and Spark SQL queries to generate data for the D3 javascript and Matplotlib libraries. It is based on examples that are referenced in the code comments.

This section can be skipped if you are more interested in the results than the underlying technology.

In [1]:
 
# see http://stackoverflow.com/questions/29783520/create-pyspark-profile-for-ipython
import findspark
findspark.init()
import pyspark
sc = pyspark.SparkContext(appName="ProntoChallenge")
In [2]:
 
# see http://spark.apache.org/docs/latest/sql-programming-guide.html#starting-point-sqlcontext
from pyspark.sql import SQLContext
sqlContext = SQLContext(sc)
In [3]:
 
import pandas as pd
import numpy as np
 
Load the data.  Pandas currently is currently more convenient than Spark at reading csv files of the format provided by Pronto, so we'll read them as Pandas dataframes.

Load the data. Pandas currently is currently more convenient than Spark at reading csv files of the format provided by Pronto, so we'll read them as Pandas dataframes.

In [4]:
 
station_df = pd.read_csv('2015_station_data.csv')
trip_df = pd.read_csv('2015_trip_data.csv')
trip_df.dropna()
weather_df = pd.read_csv('2015_weather_data.csv')
# it's bigger dataset, and not used in this example.
#status_df = pd.read_csv('2015_status_data.csv')
 
## Visualization of Inter- and Intra-region Trips
​
This notebook does not do automated clustering on the data.  But as a quick alternative, it turns out the station ids have a region label in them.  The following lines make that an explicit column in the station table.

Visualization of Inter- and Intra-region Trips

This notebook does not do automated clustering on the data. But as a quick alternative, it turns out the station ids have a region label in them. The following lines make that an explicit column in the station table.

In [5]:
 
trip_df.loc[:,'from_region'] = trip_df.loc[:,'from_station_id'].map(lambda r: str(r).split('-')[0])
trip_df.loc[:,'to_region'] = trip_df.loc[:,'to_station_id'].map(lambda r: str(r).split('-')[0])
station_df.loc[:,'region'] = station_df.loc[:,'terminal'].map(lambda r: str(r).split('-')[0])
 
The Pandas library can be used alone, but in some situations the SQL language is more useful.  The Spark library converts easily between Pandas and Spark SQL Dataframes.  The following are SQL versions of the tables.

The Pandas library can be used alone, but in some situations the SQL language is more useful. The Spark library converts easily between Pandas and Spark SQL Dataframes. The following are SQL versions of the tables.

In [6]:
 
stations_sdf = sqlContext.createDataFrame(station_df)
stations_sdf.registerTempTable('stations')
In [7]:
 
trip_sdf = sqlContext.createDataFrame(trip_df)
trip_sdf.registerTempTable('trips')
 
The SQL tables can now be used to group and count the trips by region labels.  Pandas could be used to express this query as well, though the specific query would be different.

The SQL tables can now be used to group and count the trips by region labels. Pandas could be used to express this query as well, though the specific query would be different.

In [8]:
 
regions_sql = """
    SELECT from_region, to_region, count(*) as trips
    FROM trips
    GROUP BY from_region, to_region
"""
region_result = sqlContext.sql(regions_sql)
 
One benefit of Pandas dataframes is they display automatically as a table in Jupyter notebooks.
This is just the first five lines of the region result.

One benefit of Pandas dataframes is they display automatically as a table in Jupyter notebooks. This is just the first five lines of the region result.

In [9]:
 
region_result.toPandas()[:5]
Out[9]:
from_region to_region trips
0 DPD WF 35
1 CH UW 388
2 UW UW 2828
3 CH DPD 411
4 UW DPD 212
 
This table is easier to interpret as a cross-tabulation.
The cross-tabulation will also make it easier to display as a chart later.

This table is easier to interpret as a cross-tabulation. The cross-tabulation will also make it easier to display as a chart later.

In [10]:
 
region_to_from_df = region_result.toPandas()
region_crosstab_df = pd.crosstab(trip_df.from_region, trip_df.to_region)
region_crosstab_df
Out[10]:
to_region BT CBD CD CH DPD EL FH ID PS Pronto shop SLU UD UW WF
from_region
BT 2200 3587 5 728 842 627 49 180 982 0 4809 111 50 1503
CBD 2974 3195 27 1341 184 305 159 608 2174 2 4175 75 31 1544
CD 8 73 13 116 2 1 36 57 82 0 17 3 0 32
CH 2150 6493 236 11698 411 1133 1284 677 1502 4 8105 594 388 355
DPD 524 165 3 112 316 282 12 4 79 0 1212 412 235 35
EL 559 244 1 199 284 1515 22 7 82 0 2473 601 472 71
FH 177 708 48 1345 30 101 231 225 389 0 568 18 10 38
ID 100 482 18 214 6 5 80 137 140 0 132 4 3 143
PS 475 1372 21 288 64 53 99 156 465 0 764 5 0 960
Pronto shop 1 0 0 0 0 0 0 0 0 0 0 0 0 0
SLU 4365 4601 13 1781 1454 2703 141 179 1038 7 14944 513 324 1340
UD 106 69 5 192 361 575 7 1 7 0 475 1984 1784 21
UW 29 77 0 195 212 601 2 1 6 0 309 1731 2828 29
WF 833 1339 16 47 32 27 17 93 1125 0 763 28 23 5560
 
For use with the javascript D3 visualization library, one last format conversion to JSON is helpful.

For use with the javascript D3 visualization library, one last format conversion to JSON is helpful.

In [11]:
 
import json
​
# see http://stackoverflow.com/questions/3488934/simplejson-and-numpy-array
class NumpyAwareJSONEncoder(json.JSONEncoder):
    def default(self, obj):
        if isinstance(obj, np.ndarray) and obj.ndim == 1:
            return obj.tolist()
​
region_json = json.dumps(region_crosstab_df.as_matrix().tolist(), cls=NumpyAwareJSONEncoder)
region_labels = list(region_crosstab_df.columns.values)
 
The following section is based on a demonstration by Ariel Keselman on integrating the D3 javascript library with IPython/Jupyter notebooks, specifically with chord charts.
​
Note that in the current version of this notebook, the json matrix and region labels have been copied to the javascript template by hand.

The following section is based on a demonstration by Ariel Keselman on integrating the D3 javascript library with IPython/Jupyter notebooks, specifically with chord charts.

Note that in the current version of this notebook, the json matrix and region labels have been copied to the javascript template by hand.

In [12]:
 
# This section is modified from https://github.com/skariel/IPython_d3_js_demo/blob/master/d3_js_demo.ipynb
from IPython.display import IFrame
import re
​
def replace_all(txt,d):
    rep = dict((re.escape('{'+k+'}'), str(v)) for k, v in d.items())
    pattern = re.compile("|".join(rep.keys()))
    return pattern.sub(lambda m: rep[re.escape(m.group(0))], txt)    
​
count=0
def serve_html(s,w,h):
    import os
    global count
    count+=1
    fn= '__tmp'+str(os.getpid())+'_'+str(count)+'.html'
    with open(fn,'w') as f:
        f.write(s)
    fn = '__tmp22236_1.html'
    return IFrame('files/'+fn,w,h)
​
def chord_chart(w=800,
                h=800,
                ball_count=150,
                rad_min=2,
                rad_fac=11,
                color_count=14):
    d={
       'width'      :w,
       'height'     :h,
       'ball_count' :ball_count,
       'rad_min'    :rad_min,
       'rad_fac'    :rad_fac,
       'color_count':color_count
       }
    with open('chord.template','r') as f:
        s=f.read()
    s= replace_all(s,d)        
    return serve_html(s,w+30,h+30)
 
## Note: unfortunately the chart will not render in a static notebook, such as on github.

Note: unfortunately the chart will not render in a static notebook, such as on github.

 
The arcs around the perimeter represent the region labels for the stations, and their size represents the number of trips.  The paths between regions represent all the intra-region trips.  The width of the path where it touches a region arc is the number of trips originating in that region.  If you hover over a region trips in other regions will be dimmed.
​
At first glance this chart has a few noticable features.  The large number of intra-region trips stands out, though several smaller regions are exceptions (First Hill, Belltown and Central Business District.  Some regions, such as UW and UD are tightly connected only to each other.  Capitol Hill does originate more trips to South Lake Union, the Central Business District and Belltown than return, but the effect doesn't dominate the chart.  Pioneer Square has very even connections to other regions, which may be a function of the Second Avenue protected bike lane.

The arcs around the perimeter represent the region labels for the stations, and their size represents the number of trips. The paths between regions represent all the intra-region trips. The width of the path where it touches a region arc is the number of trips originating in that region. If you hover over a region trips in other regions will be dimmed.

At first glance this chart has a few noticable features. The large number of intra-region trips stands out, though several smaller regions are exceptions (First Hill, Belltown and Central Business District. Some regions, such as UW and UD are tightly connected only to each other. Capitol Hill does originate more trips to South Lake Union, the Central Business District and Belltown than return, but the effect doesn't dominate the chart. Pioneer Square has very even connections to other regions, which may be a function of the Second Avenue protected bike lane.

In [13]:
 
chord_chart(ball_count=30, color_count=14, rad_fac=10, rad_min=3, w=800, h=800)
Out[13]:
 
The thing to remember is the size of the regions in the chart above is based on the number of trips.  Another query can show how many stations are in each region.  This query also assigns random colors (different from the chart above) for use later.

The thing to remember is the size of the regions in the chart above is based on the number of trips. Another query can show how many stations are in each region. This query also assigns random colors (different from the chart above) for use later.

In [14]:
 
stations_per_region_sql = """
    SELECT region, count(*) as n_stations
    FROM stations
    GROUP BY region
"""
stations_per_region = sqlContext.sql(stations_per_region_sql).toPandas()
​
import random
random.seed(20151115)
stations_per_region.loc[:,'color'] = stations_per_region.iloc[:,0].map(
    lambda n: '#%02X%02X%02X' % (random.randint(0,255),
                                 random.randint(0,255),
                                 random.randint(0,255)))
stations_per_region
Out[14]:
region n_stations color
0 UD 4 #C9BE53
1 UW 6 #7557AA
2 DPD 2 #D84B0A
3 BT 4 #A8C00D
4 ID 1 #3A949E
5 CBD 6 #E861ED
6 CD 1 #CB3C65
7 CH 10 #D64025
8 PS 2 #C1F771
9 WF 2 #7F7B14
10 EL 3 #ED8E33
11 SLU 11 #1D487F
12 FH 2 #B318E2
 
This shows that South Lake Union is a group of only ten stations.  It's the same for Capitol Hill.  UW and the University District combined are also 10 stations.  Pioneer square has only two stations, but they seem to be serving a distinct purpose, as are the four BT stations and the two waterfront stations.
​
My immediate concern is whether we currently have three network fragments (UW/UD, Capitol Hill and SLU/CBD/BT) rather than one dense network.  The data does have enough information to check how our network is performing compared to some existing networks in the US. This will be the topic of the next chart.

This shows that South Lake Union is a group of only ten stations. It's the same for Capitol Hill. UW and the University District combined are also 10 stations. Pioneer square has only two stations, but they seem to be serving a distinct purpose, as are the four BT stations and the two waterfront stations.

My immediate concern is whether we currently have three network fragments (UW/UD, Capitol Hill and SLU/CBD/BT) rather than one dense network. The data does have enough information to check how our network is performing compared to some existing networks in the US. This will be the topic of the next chart.

 
## 15 Minute Destinations
​
Analysis of other bike share networks shows that usage statistics rise non-linearly with station density.  http://nacto.org/wp-content/uploads/2015/09/NACTO_Walkable-Station-Spacing-Is-Key-For-Bike-Share_Sc.pdf
​
The following query simply counts station pairs
that have had a trip of less than 15 minutes.
They are then grouped and sorted from most destinations to least.

15 Minute Destinations

Analysis of other bike share networks shows that usage statistics rise non-linearly with station density. http://nacto.org/wp-content/uploads/2015/09/NACTO_Walkable-Station-Spacing-Is-Key-For-Bike-Share_Sc.pdf

The following query simply counts station pairs that have had a trip of less than 15 minutes. They are then grouped and sorted from most destinations to least.

In [15]:
 
stations_within_15_min_sql = """
    SELECT DISTINCT from_station_id as station_id,
                    from_station_name as station_name,
                    count(DISTINCT to_station_id) as destinations
    FROM trips
    WHERE tripduration <= (15*60)
    AND from_station_id != to_station_id
    GROUP BY from_station_id, from_station_name
    ORDER BY destinations DESC
"""
n_destinations_df = sqlContext.sql(stations_within_15_min_sql).toPandas()
n_destinations_df
Out[15]:
station_id station_name destinations
0 CH-05 15th Ave E & E Thomas St 44
1 CH-02 E Harrison St & Broadway Ave E 43
2 CH-03 Summit Ave E & E Republican St 43
3 CH-01 Summit Ave & E Denny Way 42
4 SLU-01 REI / Yale Ave N & John St 42
5 CH-08 Cal Anderson Park / 11th Ave & Pine St 42
6 CH-15 12th Ave & E Mercer St 42
7 CH-12 Bellevue Ave & E Pine St 41
8 CH-07 E Pine St & 16th Ave 41
9 SLU-16 Pine St & 9th Ave 40
10 CBD-03 7th Ave & Union St 40
11 SLU-15 Westlake Ave & 6th Ave 39
12 CH-06 12th Ave & E Denny Way 39
13 CH-09 Harvard Ave & E Pine St 39
14 CBD-13 2nd Ave & Pine St 39
15 SLU-07 PATH / 9th Ave & Westlake Ave 38
16 SLU-20 Terry Ave & Stewart St 37
17 CBD-06 2nd Ave & Spring St 37
18 BT-05 2nd Ave & Blanchard St 36
19 FH-01 Frye Art Museum / Terry Ave & Columbia St 36
20 CBD-04 Union St & 4th Ave 35
21 BT-04 6th Ave & Blanchard St 35
22 SLU-18 Dexter Ave & Denny Way 34
23 SLU-04 Republican St & Westlake Ave N 34
24 CBD-07 City Hall / 4th Ave & James St 33
25 DPD-01 9th Ave N & Mercer St 33
26 FH-04 Seattle University / E Columbia St & 12th Ave 32
27 BT-03 2nd Ave & Vine St 32
28 BT-01 3rd Ave & Broad St 31
29 CBD-05 1st Ave & Marion St 29
30 SLU-02 Dexter Ave N & Aloha St 29
31 SLU-19 Key Arena / 1st Ave N & Harrison St 29
32 PS-05 King Street Station Plaza / 2nd Ave Extension ... 28
33 PS-04 Occidental Park / Occidental Ave S & S Washing... 28
34 EL-03 E Blaine St & Fairview Ave E 27
35 ID-04 6th Ave S & S King St 27
36 SLU-17 Lake Union Park / Valley St & Boren Ave N 26
37 EL-01 Fred Hutchinson Cancer Research Center / Fairv... 25
38 CD-01 12th Ave & E Yesler Way 23
39 WF-01 Pier 69 / Alaskan Way & Clay St 23
40 WF-04 Seattle Aquarium / Alaskan Way S & Elliott Bay... 19
41 EL-05 Eastlake Ave E & E Allison St 18
42 UD-02 NE 42nd St & University Way NE 16
43 UW-04 15th Ave NE & NE 40th St 14
44 UD-04 12th Ave & NE Campus Pkwy 14
45 UW-02 Burke Museum / E Stevens Way NE & Memorial Way NE 13
46 SLU-21 Mercer St & 9th Ave N 13
47 UD-07 NE 47th St & 12th Ave NE 13
48 UW-10 UW Magnuson Health Sciences Center Rotunda / C... 12
49 UW-07 UW Intramural Activities Building 11
50 UW-06 UW Engineering Library / E Stevens Way NE & Je... 11
51 UD-01 Burke-Gilman Trail / NE Blakeley St & 24th Ave NE 11
52 UW-01 UW McCarty Hall / Whitman Ct 10
53 DPD-03 Children's Hospital / Sandpoint Way NE & 40th ... 6
54 Pronto shop Pronto shop 1
 
This shows that there are 10 stations that are reasonably well connected (40+ destinations).  The two busy waterfront stations actually only have a mediocre number of destinations (19 and 23).  The University stations have very low connectivity and Children's Hospital is an outlier at the edge of the network.
​
The graphs in the NACTO paper suggest that stations with 40 destinations could expect to have around 10-30 departures per day.  This can also be checked.

This shows that there are 10 stations that are reasonably well connected (40+ destinations). The two busy waterfront stations actually only have a mediocre number of destinations (19 and 23). The University stations have very low connectivity and Children's Hospital is an outlier at the edge of the network.

The graphs in the NACTO paper suggest that stations with 40 destinations could expect to have around 10-30 departures per day. This can also be checked.

In [17]:
 
departures_per_day_sql = """
    SELECT from_station_id as station_id, count(*)/365 as departures
    FROM trips
    GROUP BY from_station_id
    ORDER BY departures DESC
"""
arrivals_per_day_sql = """
    SELECT to_station_id as station_id, count(*)/365 as arrivals
    FROM trips
    GROUP BY to_station_id
    ORDER BY arrivals DESC
"""
departures_df = sqlContext.sql(departures_per_day_sql).toPandas()
arrivals_df = sqlContext.sql(arrivals_per_day_sql).toPandas()
n_trips_df = arrivals_df.merge(departures_df).merge(n_destinations_df)
n_trips_df.loc[:,'trips_to_from'] = n_trips_df.loc[:,'arrivals'] + n_trips_df.loc[:,'departures']
​
# merge in region, stations_per_region and region color
n_trips_df.loc[:,'region'] = n_trips_df.loc[:,'station_id'].map(lambda r: str(r).split('-')[0])
n_trips_df = n_trips_df.merge(stations_per_region.loc[:,['region','color']], on='region')
​
n_trips_df
Out[17]:
station_id arrivals departures station_name destinations trips_to_from region color
0 WF-01 19.758904 18.471233 Pier 69 / Alaskan Way & Clay St 23 38.230137 WF #7F7B14
1 WF-04 12.106849 8.660274 Seattle Aquarium / Alaskan Way S & Elliott Bay... 19 20.767123 WF #7F7B14
2 CBD-13 19.695890 14.753425 2nd Ave & Pine St 39 34.449315 CBD #E861ED
3 CBD-06 12.512329 8.246575 2nd Ave & Spring St 37 20.758904 CBD #E861ED
4 CBD-05 11.241096 7.936986 1st Ave & Marion St 29 19.178082 CBD #E861ED
5 CBD-03 10.846575 8.147945 7th Ave & Union St 40 18.994521 CBD #E861ED
6 CBD-07 5.295890 5.035616 City Hall / 4th Ave & James St 33 10.331507 CBD #E861ED
7 CBD-04 1.791781 1.890411 Union St & 4th Ave 35 3.682192 CBD #E861ED
8 BT-01 15.890411 16.123288 3rd Ave & Broad St 31 32.013699 BT #A8C00D
9 BT-05 9.476712 9.227397 2nd Ave & Blanchard St 36 18.704110 BT #A8C00D
10 BT-03 9.276712 11.504110 2nd Ave & Vine St 32 20.780822 BT #A8C00D
11 BT-04 5.084932 6.084932 6th Ave & Blanchard St 35 11.169863 BT #A8C00D
12 SLU-07 14.767123 10.213699 PATH / 9th Ave & Westlake Ave 38 24.980822 SLU #1D487F
13 SLU-15 14.597260 13.715068 Westlake Ave & 6th Ave 39 28.312329 SLU #1D487F
14 SLU-16 13.265753 9.624658 Pine St & 9th Ave 40 22.890411 SLU #1D487F
15 SLU-04 12.821918 8.279452 Republican St & Westlake Ave N 34 21.101370 SLU #1D487F
16 SLU-01 12.079452 12.194521 REI / Yale Ave N & John St 42 24.273973 SLU #1D487F
17 SLU-19 11.235616 10.315068 Key Arena / 1st Ave N & Harrison St 29 21.550685 SLU #1D487F
18 SLU-02 9.830137 8.972603 Dexter Ave N & Aloha St 29 18.802740 SLU #1D487F
19 SLU-17 8.849315 8.243836 Lake Union Park / Valley St & Boren Ave N 26 17.093151 SLU #1D487F
20 SLU-18 5.610959 7.038356 Dexter Ave & Denny Way 34 12.649315 SLU #1D487F
21 SLU-20 2.945205 2.704110 Terry Ave & Stewart St 37 5.649315 SLU #1D487F
22 SLU-21 0.150685 0.213699 Mercer St & 9th Ave N 13 0.364384 SLU #1D487F
23 PS-04 12.928767 7.701370 Occidental Park / Occidental Ave S & S Washing... 28 20.630137 PS #C1F771
24 PS-05 9.183562 5.235616 King Street Station Plaza / 2nd Ave Extension ... 28 14.419178 PS #C1F771
25 DPD-01 9.446575 7.002740 9th Ave N & Mercer St 33 16.449315 DPD #D84B0A
26 DPD-03 2.054795 2.287671 Children's Hospital / Sandpoint Way NE & 40th ... 6 4.342466 DPD #D84B0A
27 EL-03 9.389041 7.293151 E Blaine St & Fairview Ave E 27 16.682192 EL #ED8E33
28 EL-05 6.621918 5.389041 Eastlake Ave E & E Allison St 18 12.010959 EL #ED8E33
29 EL-01 5.709589 5.208219 Fred Hutchinson Cancer Research Center / Fairv... 25 10.917808 EL #ED8E33
30 CH-08 8.284932 13.367123 Cal Anderson Park / 11th Ave & Pine St 42 21.652055 CH #D64025
31 CH-02 8.172603 13.123288 E Harrison St & Broadway Ave E 43 21.295890 CH #D64025
32 CH-09 5.254795 8.016438 Harvard Ave & E Pine St 39 13.271233 CH #D64025
33 CH-03 5.208219 8.660274 Summit Ave E & E Republican St 43 13.868493 CH #D64025
34 CH-12 5.153425 7.021918 Bellevue Ave & E Pine St 41 12.175342 CH #D64025
35 CH-01 5.087671 8.863014 Summit Ave & E Denny Way 42 13.950685 CH #D64025
36 CH-07 4.539726 14.219178 E Pine St & 16th Ave 41 18.758904 CH #D64025
37 CH-05 3.356164 10.347945 15th Ave E & E Thomas St 44 13.704110 CH #D64025
38 CH-15 3.183562 7.778082 12th Ave & E Mercer St 42 10.961644 CH #D64025
39 CH-06 1.775342 4.575342 12th Ave & E Denny Way 39 6.350685 CH #D64025
40 UD-04 6.476712 5.216438 12th Ave & NE Campus Pkwy 14 11.693151 UD #C9BE53
41 UD-01 5.745205 4.841096 Burke-Gilman Trail / NE Blakeley St & 24th Ave NE 11 10.586301 UD #C9BE53
42 UD-07 2.693151 3.523288 NE 47th St & 12th Ave NE 13 6.216438 UD #C9BE53
43 UD-02 1.739726 1.726027 NE 42nd St & University Way NE 16 3.465753 UD #C9BE53
44 ID-04 6.369863 4.010959 6th Ave S & S King St 27 10.380822 ID #3A949E
45 UW-04 4.205479 3.761644 15th Ave NE & NE 40th St 14 7.967123 UW #7557AA
46 UW-06 3.468493 3.758904 UW Engineering Library / E Stevens Way NE & Je... 11 7.227397 UW #7557AA
47 UW-10 3.153425 2.304110 UW Magnuson Health Sciences Center Rotunda / C... 12 5.457534 UW #7557AA
48 UW-07 2.769863 2.394521 UW Intramural Activities Building 11 5.164384 UW #7557AA
49 UW-02 2.030137 2.972603 Burke Museum / E Stevens Way NE & Memorial Way NE 13 5.002740 UW #7557AA
50 UW-01 1.216438 1.301370 UW McCarty Hall / Whitman Ct 10 2.517808 UW #7557AA
51 FH-04 3.758904 5.104110 Seattle University / E Columbia St & 12th Ave 32 8.863014 FH #B318E2
52 FH-01 2.101370 5.547945 Frye Art Museum / Terry Ave & Columbia St 36 7.649315 FH #B318E2
53 CD-01 1.112329 1.205479 12th Ave & E Yesler Way 23 2.317808 CD #CB3C65
In [18]:
 
%matplotlib inline
import matplotlib.pyplot as plt
​
_, rides = plt.subplots()
rides.grid(color='grey',
           linestyle='dotted')
​
scatter = rides.scatter(n_trips_df.loc[:,'destinations'],
                        n_trips_df.loc[:,'trips_to_from'],
                        c=n_trips_df.loc[:,'color'],
                        s = n_trips_df.loc[:,'destinations'])
​
rides.set_xlabel('other docks within 15 minute ride')
rides.set_ylabel('rides per day to/from dock')
rides.set_title("15 Minute Rides", size=14);
 
This plot looks similar to the plots in the NACTO paper.  The stations with the most destinations have between 10 and 30 rides per day.  The more remote stations appear to have fewer rides.  Because this notebook did not correct for stations that were added during the year, a few stations such as the new station at 9th and Mercer do not have an accurate rides per day number. 
​
Note also that the region colors in this plot do not correspond to the colors in the chord chart above, and there is no color/region key.  Still it is possible to see that the different regions have slightly different usage characteristics beyond station density and region size.

This plot looks similar to the plots in the NACTO paper. The stations with the most destinations have between 10 and 30 rides per day. The more remote stations appear to have fewer rides. Because this notebook did not correct for stations that were added during the year, a few stations such as the new station at 9th and Mercer do not have an accurate rides per day number.

Note also that the region colors in this plot do not correspond to the colors in the chord chart above, and there is no color/region key. Still it is possible to see that the different regions have slightly different usage characteristics beyond station density and region size.

 
## Conclusions and Next Steps
​
This notebook is intended to provides some quick insight into Pronto performance after a year of operation and toin demonstrate some useful technologies for continuing investigation.
​
The inter-region trip chart seems to show that there is some fragmentation and hill effects in the network, but they are not out of scale when compared to intra-region trips in the larger regions and the well-connected central business district, Belltown and even pioneer square regions.
​
The 15 minute rides chart shows that our network has similar performance to other bike share systems, but with the caveat that less dense and peripheral stations do not have enough destinations see high usage.
​
The Pandas, Spark and D3 libraries make it easy to manipulate and visualize this data.  Similar charts can be generated by demographics, time of day or year with slight modifications to the query.  It would be interesting to see the charts with automated region clusters rather than the region labels from the station table.
​

Conclusions and Next Steps

This notebook is intended to provides some quick insight into Pronto performance after a year of operation and toin demonstrate some useful technologies for continuing investigation.

The inter-region trip chart seems to show that there is some fragmentation and hill effects in the network, but they are not out of scale when compared to intra-region trips in the larger regions and the well-connected central business district, Belltown and even pioneer square regions.

The 15 minute rides chart shows that our network has similar performance to other bike share systems, but with the caveat that less dense and peripheral stations do not have enough destinations see high usage.

The Pandas, Spark and D3 libraries make it easy to manipulate and visualize this data. Similar charts can be generated by demographics, time of day or year with slight modifications to the query. It would be interesting to see the charts with automated region clusters rather than the region labels from the station table.