Naive Bayes

This is a Naive Bayes classifier model written in Rust. It is part of the Study Material Sharing Website, where it is used to classify different categories of uploaded materials.

How It Works

The model provides two endpoints: train and predict. These endpoints allow the model to function seamlessly within a Docker container. The endpoints are implemented using Actix Web.

The model saves its state to a .bin file in the /model folder. Each training session is saved as a separate version, making it easy to roll back to a previous model state.

Technologies Used

• Rust (2021 Edition)
• Actix Web (4.9)
• Bincode (1.3)

Endpoints

Train Endpoint

URL: localhost:8080/train
Method: POST

The train endpoint expects a JSON request body with the following fields:

1. text - The training text.
2. class - The category to which the text should be classified.

Response:
Returns either Model successfully trained and data saved! or an error message.

Example Request:

{
    "text": "This is an example text.",
    "class": "example"
}

Predict Endpoint

URL: localhost:8080/predict
Method: POST

The predict endpoint expects a JSON request body with the following field:

1. text - The text whose class is to be predicted.

Response:
Returns the predicted class or an error message.

Example Request:

{
    "text": "This is another example text."
}

Saving and Loading

The model automatically saves its state after each training session. The primary save file is model.bin, located in the /model folder. This file represents the latest state of the model and is overwritten with each save.

File Structure:

any-directory/
├── NaiveBayes
├── model/
|   ├── model.bin
|   └── versions/
|       ├── model--<class>--<date>.bin
|       └── ...

Initialization:

When initialized, the model searches for a /model folder in the same directory as the executable. If the folder does not exist, it will be created automatically. The binary file /model/model.bin is loaded during initialization.

Rolling Back:

To use an older version of the model, rename the desired version file to model.bin and place it in the /model folder, replacing the current file. Note that the version files are optional and can be deleted at any time.

Training the Model

The recommended way to train the model is locally. Follow these steps:

1. Build the Naive Bayes binary using:

   cargo build --release

2. Run the binary from the /target/release/ directory. This will start the server and open the endpoints for training.

Post-Training:

After training, the /model/model.bin file is the only important artifact. You can delete everything else, including the compiled binary.

To deploy the trained model, place the model.bin file into a Docker container. This ensures the model is ready for usage without further configuration. The file should be put in a /model folder alongside the NaiveBayes binary.