PyCaret 2.3.6, incredible update; Dashboard and EDA functions

PyCaret is a useful auto ML python library because we can deploy machine learning models with low codes. We can also perform preprocessing, compare models, and tune hyperparameters, of course with low codes.

Recently, PyCaret version 2.3.6 was released. This is big news because several new wonderful functions were implemented in this release version! The details are described in this article written by PyCaret creator.

In this article, we will check the summary of this release. And, the three new features will be introduced.

New Features

• Dashboard: interactive dashborad for a trained model.
• EDA: Explonatory Data Analysis
• Convert Model: converting a trained model from python into other programing language, such as C, Java, Go, JavaScript, Visual Basic, C#, PowerShell, R, PHP, Dart, Haskell, Ruby, F#.
• Fairness
• Web API based on FastAPI
• Create Dockerfile and requirements file for a API
• Web Application based on Gradio
• Monitor Data Drift
• Optimize Threshold for classification
• New Documentation

As seeing the above new feature list, PyCaret is evolving dramatically!

In the following, we will introduce some of the new functions using normal regression analysis as an example.

Installation

If you have NOT installed PyCaret yet, you can easily install it by the following command. Note that specify the PyCaret version!

$pip install pycaret==2.3.6

From here, the sample code in this post is supposed to run on Jupyter Notebook.

Import Libraries

In advance, we load all the modules for regression analysis of PyCaret.

from pycaret.regression import *

Dataset

We use the diamond dataset for regression analysis.

# load dataset
from pycaret.datasets import get_data

df = get_data('diamond')

Set up the environment by the “setup()” function

PyCaret needs to initialize an environment by the “setup()” function. Conveniently, PyCaret infers the data type of the variables in the dataset.

Arguments of setup() are the dataset as Pandas DataFrame, the target-column name, and the “session_id”. The “session_id” equals a random seed.

s = setup(df, target='Price', session_id = 20220121)

Create a Model

Due to the simplicity of the technique and the interpretability of the model, we will adopt lr(Linear Regression) for the models that will be used below.

We can create the selected model by create_model() with the argument of “lr”. Another argument of “fold” is the number of cross-validation. “fold = 4” indicates we split the dataset into four and train the model in each dataset separately.

lr = create_model("lr", fold=4)

Introduction of New Features of PyCaret 2.3.6

From here, we will introduce two of the new features.

Dashboard

With this new function, we can create a dashboard for a trained model.

The dashboard function is implemented by ExplainerDashboard, we need the “explainerdashboard” library. We can install it with the pip command.

$pip install explainerdashboard

Then, we can create a dashboard.

dashboard(model)

Parts of the dashboard screen are introduced in the figure below.

EDA(Exploratory Data Analysis)

This new function requires the “autoviz” library.

$pip install autoviz

Just 1 line code. We can perform the EDA.

eda()

Summary

We have seen the new features of PyCaret 2.3.6.

In this article, we saw the Dashboard and EDA function.

Just 1 line.

We can create a dashboard and perform the EDA of a trained model. Wouldn’t it be great? If you sympathize with it, please give it a try.

The author hopes this blog helps readers a little.