Hi! Glad to see you made it to the Quickstart guide! In this guide you are introduced to how CARTOframes can be used by data scientists in spatial analysis workflows. Using simulated Starbucks revenue data, this guide walks through some common steps a data scientist takes to answer the following question: which stores are performing better than others?
Before you get started, we encourage you to have CARTOframes and Python 3 installed so you can get a feel for the library by using it:
In [1]:
pip install cartoframes
For additional ways to install CARTOframes, check out the Installation Guide.
Let's say you are a data scientist working for Starbucks and you want to better understand why some stores in Brooklyn, New York, perform better than others.
To begin, let's outline a workflow:
Let's get started!
We are going to use a dataset that contains information about the location of Starbucks and each store's annual revenue.
As a first exploratory step, read it into a Jupyter Notebook using pandas.
In [2]:
import pandas as pd
stores_df = pd.read_csv('http://libs.cartocdn.com/cartoframes/files/starbucks_brooklyn.csv')
stores_df.head()
Out[2]:
To display your stores as points on a map, you first have to convert the address column into geometries. This process is called geocoding and CARTO provides a straightforward way to do it (you can learn more about it in the Location Data Services guide).
In order to geocode, you have to set your CARTO credentials. If you aren't sure about your API key, check the Authentication guide to learn how to get it. In case you want to see the geocoded result, without being logged in, you can get it here.
Note: If you don't have an account yet, you can get a trial, or a free account if you are a student, by signing up here.
In [3]:
from cartoframes.auth import set_default_credentials
set_default_credentials('creds.json')
Now that your credentials are set, we are ready to geocode the dataframe. The resulting data will be a GeoDataFrame.
In [4]:
from cartoframes.data.services import Geocoding
stores_gdf, _ = Geocoding().geocode(stores_df, street='address')
stores_gdf.head()
Out[4]:
Done! Now that the stores are geocoded, you will notice a new column named geometry has been added. This column stores the geographic location of each store and it's used to plot each location on the map.
You can quickly visualize your geocoded dataframe using the Map and Layer classes. Check out the Visualization guide to learn more about the visualization capabilities inside of CARTOframes.
In [5]:
from cartoframes.viz import Map, Layer
Map(Layer(stores_gdf))
Out[5]:
Great! You have a map!
With the stores plotted on the map, you now have a better sense about where each one is. To continue your exploration, you want to know which stores earn the most yearly revenue. To do this, you can use the size_continuous_style visualization layer:
In [6]:
from cartoframes.viz import Map, Layer, size_continuous_style
Map(Layer(stores_gdf, size_continuous_style('revenue', size_range=[10,40]), title='Annual Revenue ($)'))
Out[6]:
Good job! By using the size continuous visualization style you can see right away where the stores with higher revenue are. By default, visualization styles also provide a popup with the mapped value and an appropriate legend.
Similar to geocoding, there is a straightforward method for creating isochrones to define areas of influence around each store. Isochrones are concentric polygons that display equally calculated levels over a given surface area measured by time.
For this analysis, let's create isochrones for each store that cover the area within a 15 minute walk.
To do this you will use the Isolines data service:
In [7]:
from cartoframes.data.services import Isolines
isochrones_gdf, _ = Isolines().isochrones(stores_gdf, [15*60], mode='walk')
isochrones_gdf.head()
Out[7]:
In [8]:
stores_map = Map([
Layer(isochrones_gdf),
Layer(stores_gdf, size_continuous_style('revenue', size_range=[10,40]), title='Annual Revenue ($)')
])
stores_map
Out[8]:
There they are! To learn more about creating isochrones and isodistances check out the Location Data Services guide.
Note: You will see how to publish a map in the last section. If you already want to publish this map, you can do it by calling
stores_map.publish('starbucks_isochrones', password=None).
Now that you have the area of influence calculated for each store, let's take a look at how to augment the result with population information to help better understand a store's average revenue per person.
Note: To be able to use the Enrichment functions you need an enterprise CARTO account with Data Observatory 2.0 enabled. Contact your CSM or contact us at sales@carto.com for more information.
First, let's find the demographic variable we need. We will use the Catalog class that can be filter by country and category. In our case, we have to look for USA demographics datasets. Let's see which public ones are available.
In [9]:
from cartoframes.data.observatory import Catalog
datasets_df = Catalog().country('usa').category('demographics').datasets.to_dataframe()
datasets_df[datasets_df['is_public_data'] == True]
Out[9]:
Nice! Let's take the first one (acs_sociodemogr_b758e778) that has aggregated data from 2013 to 2018 and check which of its variables have data about the total population.
In [10]:
from cartoframes.data.observatory import Dataset
dataset = Dataset.get('acs_sociodemogr_b758e778')
variables_df = dataset.variables.to_dataframe()
variables_df[variables_df['description'].str.contains('total population', case=False, na=False)]
Out[10]:
We can see the variable that contains the total population is the one with the slug total_pop_3cf008b3. Now we are ready to enrich our areas of influence with that variable.
In [11]:
from cartoframes.data.observatory import Variable
from cartoframes.data.observatory import Enrichment
variable = Variable.get('total_pop_3cf008b3')
isochrones_gdf = Enrichment().enrich_polygons(isochrones_gdf, [variable])
isochrones_gdf.head()
Out[11]:
Great! Let's see the result on a map:
In [12]:
from cartoframes.viz import color_continuous_style
Map([
Layer(isochrones_gdf, color_continuous_style('total_pop'), title='Total Population'),
Layer(stores_gdf, size_continuous_style('revenue', size_range=[10,40]), title='Annual Revenue ($)')
])
Out[12]:
At this stage, we could say that the store on the right performs better than others because its area of influence is the one with the lowest population but the store is not the one with lowest revenue. This insight will help us to focus on them in further analyses.
To learn more about discovering the data you want, check out the data discovery guide. To learn more about enriching your data check out the data enrichment guide.
The final step in the workflow is to share this interactive map with your colleagues so they can explore the information on their own. Let's do it!
Let's visualize them and add widgets to them so people are able to see some graphs of the information and filter it. To do this, we only have to add default_widget=True to the layers.
In [13]:
result_map = Map([
Layer(
isochrones_gdf,
color_continuous_style('total_pop', stroke_width=0, opacity=0.7),
title='Total Population',
default_widget=True
),
Layer(
stores_gdf,
size_continuous_style('revenue', size_range=[10,40], stroke_color='white'),
title='Annual Revenue ($)',
default_widget=True
)
])
result_map
Out[13]:
Cool! Now that you have a small dashboard to play with, let's publish it on CARTO so you are able to share it with anyone. To do this, you just need to call the publish method from the Map class:
In [14]:
result_map.publish('starbucks_analysis', password=None, if_exists='replace')
Out[14]:
In order to improve the performance and reduce the size of your map, we recommend to upload the data to CARTO and use the table names in the layers instead. To upload your data, you just need to call to_carto with your GeoDataFrame:
In [15]:
from cartoframes import to_carto
to_carto(stores_gdf, 'starbucks_stores', if_exists='replace')
to_carto(isochrones_gdf, 'starbucks_isochrones', if_exists='replace')
Out[15]: