Skip to content

Latest commit

 

History

History
75 lines (53 loc) · 5.14 KB

kaggle_guide.md

File metadata and controls

75 lines (53 loc) · 5.14 KB
Raw

Getting Started with Kaggle

This guide will help you get started with Kaggle, a cloud-based platform that allows you to run Python code directly in your browser, removing the need for a local setup. Kaggle offers powerful computing resources, including GPUs and TPUs, making it an excellent tool for data science, machine learning, and deep learning projects. It is especially useful for those who may not have access to high-performance hardware.

Please note, that Kaggle provides limited free usage of these resources. While you will not be billed for exceeding usage limits, access to GPUs or TPUs may be temporarily restricted if you reach your quota.

We hope you find this guide helpful! Follow the steps below to get started with Kaggle.

Prerequisites

Before you start, make sure you have:

  1. Your Personal Private AE4354-Y24 Repository: Clone the repository and create a private version by following the steps outlined here, if you have not done so already.

  2. A GitHub Personal Access Token: You’ll need a personal access token to use GitHub with Kaggle. Follow the instructions here to create one.

Getting Started

Follow these steps to set up Kaggle and start coding:

Step 1: Sign Up with Kaggle

  1. Go to the Kaggle website.
  2. Click on "Sign Up" and follow the instructions to create your account with your TUDelft email.

⚠️ Your Kaggle account must be associated with your TU Delft email address in order to join the course competition, which is exclusive to TU Delft students.

Step 2: Import a Notebook into Kaggle

  1. Download Your Notebook:

    • Find and download the Kaggle notebook from your GitHub repository to your computer. For example, it should be located under ex_2/ex_2_kaggle.ipynb for Exercise 2.

  2. Create a New Notebook in Kaggle:

    • Log in to Kaggle.
    • Click on + Create in the top-left corner of the homepage and select New Notebook.
    • Name your notebook. For example, ex_2_kaggle for Exercise 2.

  3. Import Your Notebook:

    • In your new notebook, click on File in the menu and select Import Notebook.
    • Upload the notebook file you downloaded.

  4. Verify Your Account:

    • Ensure your account is verified with a phone number. If you’re not prompted to verify, you can find the verification link in the Session options on the right side of the page.
    • Verification is required to access Kaggle’s internet and GPU/TPU features.

  5. Enable Internet Access:

    • Once verified, turn on the Internet option in the Session options on the right side.

Step 3: Load the Dataset in Kaggle

To work with the dataset in your Kaggle notebook, follow these steps:

  1. Check Dataset Availability:

    • After opening your notebook, look under Input → DATASETS on the right side of the notebook interface.
    • You should see the corresponding dataset there (e.g. polarization for Exercise 2). If it’s there, you’re ready to use it.

  2. Add the Dataset if Not Visible:

Step 4: Switching Between CPU and GPU

  1. Use CPU for Initial Setup:

    • Perform setup and exploration tasks using a CPU. In the Settings bar, select Accelerator and choose None. This will save your GPU quota for when you need it.

  2. Switch to GPU P100 for Training:

    • When you’re ready to run your training, switch to GPU P100 to speed up the process.
    • To switch, go to the Settings bar, select Accelerator, and choose GPU P100.

💡 After switching the accelerator, the runtime of the notebook will be restarted. You would need to re-execute the entire notebook from the beginning.

💡 Kaggle offers unlimited hours of CPU usage, 30 hours per week GPU usage, and 20 GB of auto-saved disk space under /kaggle/working. Use CPU for setup and GPU for training to make the most of your usage quota. For more details on Kaggle’s GPU usage limits, see the Kaggle Documentation. Ensure you monitor your resource usage to avoid hitting limits and budget the resource usage wisely among working on exercises and the competition.

Common Errors

  • Kernel Crashes: Restart the kernel by going to the Run dropdown and clicking Factory reset.
  • Dataset Not Found: Use the command !ls /kaggle/input to list available datasets and adjust your paths accordingly. See the notebooks for further details.

Well done! Your Kaggle setup is complete, and you are ready to start working on the project. Follow the instructions in your notebook to proceed.