AIML-EN: Exercise 4: Training, validation and testing sets

Exercise: Data Labeling with Label Studio

Objective:

Learn how to use Label Studio to label a dataset for a machine learning project.

Prerequisites:

- Basic understanding of data labeling and its importance in machine learning.

- Label Studio installed on your machine. Here are installing instructuons.

- We choose install Label Studio using Anaconda.

Steps:

1. Install and Start Label Studio using Anaconda:

- Download and install Anaconda from here https://www.anaconda.com/download

- Run Anaconda navigator, run anaconda_prompt and run next commands

conda create --name label-studio
conda activate label-studio
conda install psycopg2  # required for LS 1.7.2 only
pip install label-studio

Open your web browser and go to `http://localhost:8080` to access the Label Studio interface.

After that, label-studio enviroment is created and next time run the environment on label-studio play button -> Open Terminal, using command:

label-studio start

On first run, register account on Sign up link and confirm on link via recieved e-mail. Next time you can login using this account.

2. Create and setup New Project:

- Click on "Create" to start a new project.

- Name your project "Animal Classification" and provide a brief description.

- Go to Labeling Interface tab, Select "Image Classification" as the labeling interface and define the labels you want to use for classification using Add button. For example, if you are classifying animals, you might have labels like "Cat", "Dog", "Bird", etc. Then click "Save".

3. Upload Data:

- Prepare a dataset of images. You can use your own images or a sample dataset of images from any public source (e.g., CIFAR-10, ImageNet). Let's download some images of cats, dogs and birds from https://github.com/AIForVet/aiml/tree/main/train

- Click on "Go to import" and drop images you downloaded and want to label (cats, dogs, birds...).

- Then, click on Import button. Later, you can delete selected items or add new items.

4. Start Labeling:

- Click on Label All Tasks and begin labeling the images by selecting the appropriate label for each image an click on Submit button.

- Label at least 20 images to get a good amount of labeled data.

7. Export Labeled Data:

- Once you have labeled the images, click on "Export" to download the labeled data.

- Choose the export format (e.g., JSON) and save the file to your local machine.

8. Analyze the Labeled Data:

- Open the exported file and examine the labeled data.

- Discuss how this labeled data can be used to train a machine learning model.

Discussion Questions:

1. Why is data labeling important in machine learning?

2. What challenges did you face while labeling the data?

3. How can the quality of labeled data affect the performance of a machine learning model?

4. What strategies can be used to ensure high-quality data labeling?

Submission:

- Submit the exported labeled data file.

- Write a brief report (1-2 pages) answering the discussion questions.

Completion requirements:

Make a submission