How to visualise different ML models using PyCaret for optimization?

In machine learning, optimization of the results produced by models plays an important role in obtaining better results. We normally get these results in tabular form and optimizing models using such tabular results makes the procedure complex and time-consuming. Visualizing results in a good manner is very helpful in model optimization. Using the PyCaret library, we can visualize the results and optimize them. In this article, we are going to discuss how we can visualize results from different modelling procedures. The major points to be discussed in the article are listed below.

Table of contents

1. What is PyCaret?
2. Visualizing a classification model
3. Visualizing a regression model
4. Visualizing a clustering model
5. Visualizing anomaly detection

What is PyCaret?

In one of our articles, we have discussed that PyCaret is an open-source library that helps us in performing a variety of machine learning modelling automatically. This library aims to make the process of machine learning modelling simple using a low code interface. Along with such a low code interface, this library provides modules that are very efficient and low time-consuming.  

To improve the explainability and interpretability of the process this library provides various visualizations implemented using some of the modules. This library combines visualization from various famous packages like yellowbricks, autovig, plotly, etc. in this article using visualization modules from the PyCaret we are going to plot results from the different models. 

Let’s start with a classification model.

Visualizing a Classification model

Let’s start the procedure by importing data. Since we have multiple datasets in the PyCaret library, we can utilize them for practice purposes. In this article, we are going to use a heart disease dataset that has multiple values related to medical conditions that classify whether a person can have heart disease or not. Let’s import it directly from the library.  

from pycaret.datasets import get_data
data = get_data('heart_disease')

Output:

Here in the above output, we can see some of the values from the dataset. Now let’s set up a model.

from pycaret.classification import *
model1 = setup(data = data, target = 'Disease')

Output:

In the setup of the model, we have provided the name of the dataset and the name of the target variable. After optimizing the dataset, the setup module has provided information about the variables that the dataset consists of.

Let’s fit the model.

random_forest = create_model('rf')

Output:

Here we can see the information about the accuracy and AUC scores of the model. Hereafter fitting the model, the main purpose of the article comes into the picture.

Visualizing results 

One more thing that is great about the PyCaret library is that we can utilize its features for visualizing the results from the model that we have in tabular form.

Plotting the AUC scores:

plot_model(random_forest, plot = 'auc')

Output:

Here in the plot, we can see that we have a detailed visualization of AUC values from the model. There are various libraries like yellowbricks and autovig that help PyCaret to make such visualizations.

Let’s see the confusion matrix in a visualized form.

plot_model(random_forest , plot = 'confusion_matrix')

Output:

Here we can see the confusion matrix. We can also convert values under the confusion matrix in the percentage.

plot_model(random_forest , plot = 'confusion_matrix', plot_kwargs = {'percent' : True})

Output:

We can also plot results based on the training data by just passing use_train_data as True. Let’s plot the decision boundary of the model.

plot_model(random_forest, plot = 'boundary', use_train_data = True)

Output:

Feature engineering plays a crucial role in data modelling. We can check the feature importance using the following lines of codes.

plot_model(random_forest, plot = 'feature_all', use_train_data = True)

Output:

Here we have discussed the visualization of results by a classification model. We can also plot various results from a regression model.

Visualizing a Regression model

Let’s fit a regression model using the automobile dataset.

data = get_data('automobile')
from pycaret.regression import *
model2 = setup(data = data, target = 'price')
random_forest_r = create_model('rf')

Output:

Here we can see the results from a random forest regression in a tabular form. Let’s plot these results.

Plotting learning curve   

plot_model(random_forest_r, plot = 'learning', use_train_data = True)

Output:

Let’s plot the errors and residual values from the model.

Error plot

plot_model(random_forest_r, plot = 'error', use_train_data = True)

Output:

Residual plot

plot_model(random_forest_r, plot = 'residuals', use_train_data = True)

Output:

Here we have seen some of the visualizations of results of a regression model. We can also plot results from the cluster modelling.

Visualizing a clustering model

For clustering, we are using the mice data and we will be fitting the k means model on the data.

data = get_data('mice')
from pycaret.clustering import *
 
model3 = setup(data, normalize = True, 
                   ignore_features = ['MouseID'],
                   session_id = 123)
kmeans = create_model('kmeans')

Output:

Now we are ready to plot the results from cluster modelling.

Plotting clusters 

plot_model(kmeans, plot = 'cluster')

Output:

For this visualization, PyCaret uses the plotly package. This package provides interactive plots that are not provided here, we can find these plots here.

Let’s plot the clusters in 3D.

We can also plot the results like a silhouette and distance. Let’s plot the distance of the clusters.

plot_model(kmeans, plot = 'distance')

Output:

Here we have seen some of the results visualized from the cluster analysis. We can also visualize the results from the anomaly detection process.

Visualizing anomaly detection

Let’s fit the model on the similar data that we used for clustering.

from pycaret.anomaly import *
 
model4 = setup(data, normalize = True, 
                   ignore_features = ['MouseID'],
                   session_id = 123)
iforest = create_model('iforest')

Let’s plot the results of anomaly detection.

plot_model(iforest)

Output:

We can also plot a  umap dimensionality plot.

plot_model(iforest, plot = ‘umap’)

Output:

These plots are also generated using the plotly library. We can also see in the plot what are the anomalies in the mice dataset.

Final words

In the article, we have discussed how we can visualize the results from the model using modules of the PyCaret library. Since every plot was informative enough, we can utilize them for optimizing the models and modelling procedures so that we can improve results. 

References 

• PyCaret documentation
• Link fo the codes