Deploy your apps to a supercloud in a few clicks

This Engineering Education program is supported by Section. Instantly deploy your GitHub apps, Docker containers or K8s namespaces to a supercloud.

Try It For Free

Unsupervised Machine Learning with Pycaret

August 25, 2021

Unsupervised machine learning is a type of machine learning where the model is trained using an unlabeled dataset. The model is supposed to find the hidden patterns and information from the given dataset by itself.

Pycaret is a powerful open-source library in python that is used as a wrapper around several popular machine learning and data science libraries in building a machine learning model.

Table of content


In this tutorial, we shall use Pycaret to perform clustering, which is used in clustering countries into their respective continents. We will learn the basics of Pycaret by building an unsupervised machine learning model.

As compared to other machine learning libraries such as Scikit-learn, Pycaret is simple and easy to use due to fewer lines of code that one has to write. This increases the productivity of a developer since the developer spends less time coding.


To follow along with this tutorial, you need to have:

  1. Python installed on your machine.
  2. A basic understanding of Jupyter Notebook or Google Colab.
  3. A good understanding of Python.
  4. Latest version of Pycaret installed on your machine.

NOTE: We will be using Google Colab in this tutorial since it is a powerful notebook for building a model.

Categories of unsupervised machine learning

Unsupervised machine learning is categorized into clustering and association.


Clustering involves grouping different objects into clusters, such that objects with the most similarities will be placed in one group and those that have little or no similarity in another.

The usefulness of clustering:


Allows you to find how different data models in a dataset relate and how their relationships affect the whole dataset. Association is used in the market-based analysis, to determine which products will be bought together.

Installing Pycaret library and other dependencies

Since we are using Google Colab use the following command to install Pycaret.

!pip install pycaret

We also need to install other dependencies such as Pandas, Matplotlib and Seaborn.

1. Pandas: Pandas is used for data manipulation and analysis, it’s also used for importing and reading our CSV file dataset.

2. Matplotlib: This will be used as a plotting library and is used for visualization purposes through the drawing of statistical graphs.

3. Seaborn: Seaborn is also used for data visualization and is built on top of the Matplotlib library and builds more intuitive graphs.

We use the following command to install the above libraries.

pip install pandas matplotlib seaborn

Loading the installed packages

To use these packages, we need to load them into our machine.

import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
%matplotlib inline

Loading dataset

We need to load the dataset using Pandas, this dataset will be used to build and train our model. The dataset used is made up of four columns namely: country, latitude, longitude, and name as shown below.

Dataset snip

CSV File of our data

In the above link, you can download the CSV file of our dataset so that we can be able to load it into our machine. Name the downloaded dataset as countries_dataset.csv. We can use Pandas to read our dataset.

df = pd.read_csv("countries_dataset.csv")

Use the following command to take a look at the first 5 rows of our loaded data:


Import Pycaret

We have to import Pycaret so that we can start using it to build our model.

import pycaret.clustering as pc

Initialize setup

We initialize our setup by using the setup() method. During initialization, ignore country and name columns, since they will not affect our model so we will not need them.

Our new data will have only two columns: longitude and latitude. We also save our initialized setup into a variable named cls.

cls = pc.setup(df, ignore_features=['country', 'name'])

During this phase, we are prompted to either quit or continue the initialization process. This happens to make sure that we want to ignore the two columns country and name columns.


If we want to continue, we are supposed to type y and the initialization process will continue.

Setup completed

After the setup is successful, we can now start building the model.

Building model

To build our model, we need to know the clustering algorithm to use. We shall use the K-Means clustering algorithm due to the following reasons:

  • K-Means clustering is easy to implement
  • It is a fast algorithm compared to the other algorithms.
  • K-Means algorithm is more scalable to a huge dataset.
  • K-Means clustering gives more accurate results.

Other common algorithms used for clustering are as follows:

  1. Density-Based Spatial Clustering of Applications with Noise
  2. Balanced Iterative Reducing and Clustering using Hierarchies
  3. Agglomerative Hierarchical Clustering
  4. Mean-Shift Clustering

To start using the K-Means algorithm, we need to specify the number of clusters in which we want our countries to be grouped according to their similarities. The number of clusters specified will be 7, 7 is the number of continents in which we want the countries to be clustered.

km = pc.create_model('kmeans', num_clusters=7)

The above code snippet initializes our KMeans algorithm using the create_model() method and also specifies the number of clusters to use.

Assigning Clusters

In this section, we need to assign our 7 clusters into their respective data, this ensures that a country is placed into the right cluster, countries that belong to the same content should be placed in the same cluster. This is done using the assign_model(km) as shown below.

km_df = pc.assign_model(km)

To see how the different countries are assigned clusters, use the following command:


The output is as shown below.

Output of clusters

As shown above, the United Arab Emirates and Afghanistan have been assigned the same cluster since they belong to the same continent.

Scatter plot

A scatter plot gives a visualization of how our different countries are clustered using the following code snippets.

km_df2 = km_df.copy()

km_df2['Cluster'] = km_df2['Cluster'].str.replace('Cluster').apply(int)

The above code snippet allows us to replace the third column of the dataset with integer values so that it makes it easy when plotting a scatter plot.

After we have converted our third column from str (string) to an int (integer), we can start doing our scatter plot using the following:

plt.scatter(km_df2['longitude'], km_df2['latitude'], c=km_df2['Cluster'], cmap='rainbow')

Scatter plot

The scatter plot will give us a visual distribution of how the different countries are clustered. Our scatter plot has 7 colors to represent the number of clusters we specified earlier.

Map scatter plot onto the world map

We need to map our scatter plot on the world map so that we can have the different countries are clustered.

To perform this functionality we need first to install geopandas and descartes which enables plotting of a world map and ensure that our scatter plot is well distributed in the world map. These two packages geopandas and descartes are very important when we want to work with a world map.

pip install geopandas descartes
  • After installing, we import these packages required for plotting:
import geopandas as gpd
import descartes

We can now create our world map using the steps shown below.

World map

  • Create a world map using the geopandas package we installed earlier. We use the plot() method to plot the world map, and it has a size of 20 on the x-axis and 10 on the y-axis.
world = gpd.read_file(gpd.datasets.get_path('naturalearth_lowres'))
ax = world.plot(figsize=(20,10))
  • The output of our map will be as shown.

World map

We now need to overlap our scatter plot on top of the created world map.

Overlap scatter plot

Overlapping ensures that the different countries are clustered in the right continents, and we can also see this on the world map.

g = gpd.GeoDataFrame(km_df2,geometry=gpd.points_from_xy(km_df2['longitude'],km_df2['latitude']))

Use latitude and longitude in plotting the countries in their positions on the world map. The above code snippet will provide normal coordinates that can be used in plotting as shown below.

Plotting points

After we have generated the coordinates, We can now plot these points using the following code snippets.

fig,ax = plt.subplots(figsize=(20,10))

The output will show a scatter plot overlapped on top of our world map.

Overlap scatter plot

From the image above, we can see how different clusters are distributed over the world map.

Making Prediction

We can now test our model, by giving it a countries’ latitude and longitude coordinates and see if our model can place the country into the right cluster.

Sample latitude and longitude are: latitude: 7.946527 and longitude: -1.023194. We input these into our model and see if it can place the country into the right cluster.

sample_data = pd.DataFrame([{'latitude': 7.94652, 'longitude': -1.023194}])

We convert our sample_data into DataFrame so that our model can easily read the data frame and then use the predict_model() method to make a prediction.

clust_pred = pc.predict_model(km, data=sample_data)

To view which cluster the country belongs to, we use the following command:


Output is as shown: Prediction outcome

We can see that our model has successfully clustered the country into cluster 3 which is the right cluster. This means that we have successfully trained our model since it can now make accurate predictions.


In this tutorial, we learned how to create a clustering model. A clustering model is a type of unsupervised machine learning in which the model learns on its own without supervision.

We started with data preparation, which is an important stage in machine learning since it cleans and formats for readability. The processed data is then used in the training of our model.

In the next section, we built a clustering model, which can cluster countries into their respective continents.

Finally, we used the trained model to make a prediction, which is the goal of any machine learning model.


Peer Review Contributions by: Mercy Meave