Using Kaggle in Machine Learning Projects

You’ve probably heard of Kaggle data science competitions, but did you know that Kaggle has many other features that can help you with your next machine learning project? For people looking for datasets for their next machine learning project, Kaggle allows you to access public datasets by others and share your own datasets. For those looking to build and train their own machine learning models, Kaggle also offers an in-browser notebook environment and some free GPU hours. You can also look at other people’s public notebooks as well!

Other than the website, Kaggle also has a command-line interface (CLI) which you can use within the command line to access and download datasets.

Let’s dive right in and explore what Kaggle has to offer!

After completing this tutorial, you will learn:

• What is Kaggle?
• How you can use Kaggle as part of your machine learning pipeline
• Using Kaggle API’s Command Line Interface (CLI)

Let’s get started!

Overview

This tutorial is split into five parts; they are:

• What is Kaggle?
• Setting up Kaggle Notebooks
• Using Kaggle Notebooks with GPUs/TPUs
• Using Kaggle Datasets with Kaggle Notebooks
• Using Kaggle Datasets with Kaggle CLI tool

What Is Kaggle?

Kaggle is probably most well known for the data science competitions that it hosts, with some of them offering 5-figure prize pools and seeing hundreds of teams participating. Besides these competitions, Kaggle also allows users to publish and search for datasets, which they can use for their machine learning projects. To use these datasets, you can use Kaggle notebooks within your browser or Kaggle’s public API to download their datasets which you can then use for your machine learning projects.

In addition to that, Kaggle also offers some courses and a discussions page for you to learn more about machine learning and talk with other machine learning practitioners!

For the rest of this article, we’ll focus on how we can use Kaggle’s datasets and notebooks to help us when working on our own machine learning projects or finding new projects to work on.

Setting up Kaggle Notebooks

To get started with Kaggle Notebooks, you’ll need to create a Kaggle account either using an existing Google account or creating one using your email.

Then, go to the “Code” page.

You will then be able to see your own notebooks as well as public notebooks by others. To create your own notebook, click on New Notebook.

This will create your new notebook, which looks like a Jupyter notebook, with many similar commands and shortcuts.

You can also toggle between a notebook editor and script editor by going to File -> Editor Type.

Changing the editor type to script shows this instead:

Using Kaggle with GPUs/TPUs

Who doesn’t love free GPU time for machine learning projects? GPUs can help to massively speed up the training and inference of machine learning models, especially with deep learning models.

Kaggle comes with some free allocation of GPUs and TPUs, which you can use for your projects. At the time of this writing, the availability is 30 hours a week for GPUs and 20 hours a week for TPUs after verifying your account with a phone number.

To attach an accelerator to your notebook, go to Settings ▷ Environment ▷ Preferences.

You’ll be asked to verify your account with a phone number.

And then presented with this page which lists the amount of availability you have left and mentions that turning on GPUs will reduce the number of CPUs available, so it’s probably only a good idea when doing training/inference with neural networks.

Using Kaggle Datasets with Kaggle Notebooks

Machine learning projects are data-hungry monsters, and finding datasets for our current projects or looking for datasets to start new projects is always a chore. Luckily, Kaggle has a rich collection of datasets contributed by users and from competitions. These datasets can be a treasure trove for people looking for data for their current machine learning project or people looking for new ideas for projects.

Let’s explore how we can add these datasets to our Kaggle notebook.

First, click on Add data on the right sidebar.

A window should appear that shows you some of the publicly available datasets and gives you the option to upload your own dataset for use with your Kaggle notebook.

I’ll be using the classic titanic dataset as my example for this tutorial, which you can find by keying your search terms into the search bar on the top right of the window.

After that, the dataset is available to be used by the notebook. To access the files, take a look at the path for the file and prepend ../input/{path}. For example, the file path for the titanic dataset is:

../input/titanic/train_and_test2.csv

In the notebook, we can read the data using:

import pandas

pandas.read_csv("../input/titanic/train_and_test2.csv")

This gets us the data from the file:

Using Kaggle Datasets with Kaggle CLI Tool

Kaggle also has a public API with a CLI tool which we can use to download datasets, interact with competitions, and much more. We’ll be looking at how to set up and download Kaggle datasets using the CLI tool.

To get started, install the CLI tool using:

pip install kaggle

For Mac/Linux users, you might need:

pip install --user kaggle

Then, you’ll need to create an API token for authentication. Go to Kaggle’s webpage, click on your profile icon in the top right corner and go to Account.

From there, scroll down to Create New API Token:

This will download a kaggle.json file that you’ll use to authenticate yourself with the Kaggle CLI tool. You will have to place it in the correct location for it to work. For Linux/Mac/Unix-based operating systems, this should be placed at ~/.kaggle/kaggle.json, and for Windows users, it should be placed at C:Users<Windows-username>.kagglekaggle.json. Placing it in the wrong location and calling kaggle in the command line will give an error:

OSError: Could not find kaggle.json. Make sure it’s location in … Or use the environment method

Now, let’s get started on downloading those datasets!

To search for datasets using a search term, e.g., titanic, we can use:

kaggle datasets list -s titanic

Searching for titanic, we get:

$ kaggle datasets list -s titanic
ref                                                          title                                           size  lastUpdated          downloadCount  voteCount  usabilityRating
-----------------------------------------------------------  ---------------------------------------------  -----  -------------------  -------------  ---------  ---------------
datasets/heptapod/titanic                                    Titanic                                         11KB  2017-05-16 08:14:22          37681        739  0.7058824
datasets/azeembootwala/titanic                               Titanic                                         12KB  2017-06-05 12:14:37          13104        145  0.8235294
datasets/brendan45774/test-file                              Titanic dataset                                 11KB  2021-12-02 16:11:42          19348        251  1.0
datasets/rahulsah06/titanic                                  Titanic                                         34KB  2019-09-16 14:43:23           3619         43  0.6764706
datasets/prkukunoor/TitanicDataset                           Titanic                                        135KB  2017-01-03 22:01:13           4719         24  0.5882353
datasets/hesh97/titanicdataset-traincsv                      Titanic-Dataset (train.csv)                     22KB  2018-02-02 04:51:06          54111        377  0.4117647
datasets/fossouodonald/titaniccsv                            Titanic csv                                      1KB  2016-11-07 09:44:58           8615         50  0.5882353
datasets/broaniki/titanic                                    titanic                                        717KB  2018-01-30 04:08:45           8004        128  0.1764706
datasets/pavlofesenko/titanic-extended                       Titanic extended dataset (Kaggle + Wikipedia)  134KB  2019-03-06 09:53:24           8779        130  0.9411765
datasets/jamesleslie/titanic-cleaned-data                    Titanic: cleaned data                           36KB  2018-11-21 11:50:18           4846         53  0.7647059
datasets/kittisaks/testtitanic                               test titanic                                    22KB  2017-03-13 15:13:12           1658         32  0.64705884
datasets/yasserh/titanic-dataset                             Titanic Dataset                                 22KB  2021-12-24 14:53:06           1011         25  1.0
datasets/abhinavralhan/titanic                               titanic                                         22KB  2017-07-30 11:07:55            628         11  0.8235294
datasets/cities/titanic123                                   Titanic Dataset Analysis                        22KB  2017-02-07 23:15:54           1585         29  0.5294118
datasets/brendan45774/gender-submisson                       Titanic: all ones csv file                      942B  2021-02-12 19:18:32            459         34  0.9411765
datasets/harunshimanto/titanic-solution-for-beginners-guide  Titanic Solution for Beginner's Guide           34KB  2018-03-12 17:47:06           1444         21  0.7058824
datasets/ibrahimelsayed182/titanic-dataset                   Titanic dataset                                  6KB  2022-01-27 07:41:54            334          8  1.0
datasets/sureshbhusare/titanic-dataset-from-kaggle           Titanic DataSet from Kaggle                     33KB  2017-10-12 04:49:39           2688         27  0.4117647
datasets/shuofxz/titanic-machine-learning-from-disaster      Titanic: Machine Learning from Disaster         33KB  2017-10-15 10:05:34           3867         55  0.29411766
datasets/vinicius150987/titanic3                             The Complete Titanic Dataset                   277KB  2020-01-04 18:24:11           1459         23  0.64705884

To download the first dataset in that list, we can use:

kaggle datasets download -d heptapod/titanic --unzip

Using a Jupyter notebook to read the file, similar to the Kaggle notebook example, gives us:

Of course, some datasets are so large in size that you may not want to keep them on your own disk. Nonetheless, this is one of the free resources provided by Kaggle for your machine learning projects!

Further Reading

This section provides more resources if you’re interested in going deeper into the topic.

• Kaggle: https://www.kaggle.com
• Kaggle API documentation: https://www.kaggle.com/docs/api

Summary

In this tutorial, you learned what Kaggle is , how we can use Kaggle to get datasets, and even for some free GPU/TPU instances within Kaggle Notebooks. You’ve also seen how we can use Kaggle API’s CLI tool to download datasets for us to use in our local environments.

Specifically, you learnt:

• What is Kaggle
• How to use Kaggle notebooks along with their GPU/TPU accelerator
• How to use Kaggle datasets in Kaggle notebooks or download them using Kaggle’s CLI tool

The post Using Kaggle in Machine Learning Projects appeared first on Machine Learning Mastery.