⏳
4 mins
read time
Paris Open Data platform proposes databases of the bike rental system Velib. They propose an API to download the data in real-time. The file can be obtained using a http query to the API and saved in csv or json format.
We will use python in this post to obtain some insights about the rental systems.
import requests
response = requests.get("https://opendata.paris.fr/api/records/1.0/download/?dataset=stations-velib-disponibilites-en-temps-reel&facet=banking&facet=bonus&facet=status&facet=contract_name&rows=-1")
txt = response.text
f = open('velib.csv', 'w+')
f.write(txt)
The file contains information about the number, the name, the address and the position of the station. In addition, it also has information about the number of available bikes and the number of stands for each station. This information can be used to have a global view of the rental systems. The complete information of the file is shown below.
import pandas as pd
velibs = pd.read_csv("velib.csv", sep=";")
velibs.head()
- | number | name | address | position | banking | bonus | status | contract_name | bike_stands | available_bike_stands | available_bikes | last_update |
--- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
0 | 5035 | 05035 - BUFFON AUSTERLITZ | 1 RUE BUFFON - 75005 PARIS | 48.8431581782, 2.36374844464 | True | False | CLOSED | Paris | 24 | 0 | 0 | 2017-04-20T06:58:24+00:00 |
1 | 5104 | 05104 - BUFFON | 47 RUE BUFFON - 75005 PARIS | 48.8419161798, 2.35891268132 | True | False | CLOSED | Paris | 55 | 0 | 0 | 2017-04-20T07:56:57+00:00 |
2 | 8034 | 08034 - HAUSSMANN COURCELLES | 49 RUE DE BERRI - 75008 PARIS | 48.87481039, 2.30831406419 | True | False | OPEN | Paris | 34 | 25 | 9 | 2017-05-06T21:14:59+00:00 |
3 | 22407 | 22407 - BARBUSSE (MALAKOFF) | ROND POINT HENRI BARBUSSE - 92240 MALAKOFF | 48.8146796589, 2.28740045741 | True | True | OPEN | Paris | 25 | 20 | 4 | 2017-05-07T01:09:16+00:00 |
4 | 19116 | 19116 - JAURES VILLETTE | 180 BOULEVARD DE LA VILLETTE - 75019 PARIS | 48.8814724985, 2.37026248226 | True | False | CLOSED | Paris | 30 | 0 | 0 | 2017-05-07T01:20:16+00:00 |
We will focus only in the open stations by using the following command:
velibs = velibs[velibs.status == 'OPEN']
The number of arrondisement can be obtained from the address of station. We will focus only on the stations that are in Paris, which means that has a zip code beginning with 75. We can simply do that using the following command:
velibs['arron'] = velibs['address'].map(lambda x: int(x.split()[-2][3:]) if x.split()[-2][:2] == '75' else np.NaN)
Then we can group the number of bike stands by each arrondisement.
bike_stands_arron = velibs.groupby('arron')['bike_stands'].sum()
This information can be bind to a geojson file that contains the coordinates of each arrondisement. This information has been posted in github. We can use Folium to plot this information using the following command:
# https://github.com/codeforamerica/click_that_hood/raw/master/public/data/paris.geojson
state_geo = r'paris.json'
m = folium.Map(location=[48.856614, 2.3522219], zoom_start=13, tiles='Stamen Toner')
m.choropleth(geo_path=state_geo,
data=bike_stands_arron,
columns=['arron', 'bike_stands'],
key_on='properties.cartodb_id',
fill_color='YlGn',
fill_opacity=0.9, line_opacity=0.2,
legend_name='Number of bike stands',
highlight=1)
m
We can see graphically that the 15th arrondisement has more bikes than any other. It is mainly because it is bigger than other arrondisements.
We can normalize the number of available bikes using the total number of bikes for each station. This give information if the station is relatively empty or not. We can plot this information directly to a map using the coordinates of the station and a function that colours the station to red if the station is empty and green if the station is full.
from colour import Color
red = Color("red")
colors = list(red.range_to(Color("green").hex,10))
def red(brightness):
brightness = int(round(9 * brightness)) # convert from 0.0-1.0 to 0-255
return colors[brightness]
m = folium.Map(location=[48.856614, 2.3522219], zoom_start=13, tiles='Stamen Toner')
for k,v in velibs.iterrows():
folium.CircleMarker(location=[v.position.split(",")[0], v.position.split(",")[1]],
fill_color=red(v.available_bikes/float(v.bike_stands)).hex,
popup= str(v.available_bikes) + " / " + str(v.bike_stands),
radius=7).add_to(m)
Each point can be clicked to obtain more information about the station. Using this information over different times, we can see that during the night, there are a lot of available bikes near the outer ring of Paris and a few in the inner side of Paris, near Châtelet. On the opposite, at 8am, a lot of people take the bikes on the outer ring and they tend to go to the inner side of Paris.
Your browser doesn't support HTML5 video.
Using an scheduler cron task, we can obtain information about the evolution of the bike utilisation. For this case, download the information each hour.
Since we do not have any information about the actual number of rent bikes, we can use the number of available bikes in each station and calculate the difference in a certain time range. I take into account only when the bikes are added to the stations to avoid counting twice the utilisation of a bike (i.e when a bike is taken from a station and put in an other station we have +1, -1 for each station respectively)
During weekdays, there is a peak of utilisation at 8am and 7pm. When people is going to work. The peak of utilisation is 4552 bikes, which is around 30 % of the available bikes in the city (~15593). On weekend, the peak is at 6pm, maybe because people is coming back home.
The complete code can be found at in this notebook.
This article shows how to collect data from article shows how to analyze logs using Kibana dashboards. Fluentbit is used for injecting logs to elasticsearch, then it is connected to kibana to get some insights.
A marathon needs to be prepared for at least 3 months. This post shows my preparation training and a prediction of my time for the Metz Marathon 2017
In this post I use the javascript library Chartjs to illustrate interactively the statistics of a running activity at Santorini experience 2016
This article presents a study on predicting user-generated revenue within a digital application. Using gradient boosting models and a dataset containing user installation features and engagement metrics, the study aims to forecast revenue at day 120. The model's performance and key influencing factors are analyzed, providing insights for optimizing monetization strategies in the digital applications industry.