Dynamic Pricing Model

Dynamic Pricing is an essential aspect of modern business operations, where prices are dynamically adjusted based on various factors such as demand, seasonality, competition, and customer behavior. This project leverages machine learning to implement a robust Dynamic Pricing model, enabling businesses to optimize their pricing strategies effectively.

---

Project Overview

This project trains a machine learning model using a Gradient Boosting Regressor to predict optimal prices based on historical data. The model is managed and tracked with MLflow, which facilitates versioning, artifact storage, and model serving.

---

Features

• Data Preprocessing: Handles input data preparation, including feature extraction and splitting into training and testing sets.
• Model Training: Utilizes a Gradient Boosting Regressor for accurate price prediction.
• Model Management: Tracks experiments, logs metrics, and registers models using MLflow.
• Dynamic Model Loading: Automatically fetches and loads the latest production model for predictions.

---

Requirements

To run this project, ensure you have the following installed:

• Python 3.8+
• Required Python packages (listed in requirements.txt)
• MLflow (for model tracking and management)
• Scikit-learn (for machine learning)
• Pandas (for data manipulation)

Install dependencies using:

pip install -r requirements.txt

---

Setup and Usage

1. Data Preparation

Prepare two CSV files:

• batch_test_data.csv: test data for batch inference.
• single_test_data.csv: test data for single inference.

2. Train the Model

Train and log the model using MLflow:

This script:

1. Loads the data.
2. Splits it into training and testing sets.
3. Trains a Gradient Boosting Regressor.
4. Logs the model and metrics (e.g., MAE) in MLflow.

3. Accessing the Model

Load from Run URI

Retrieve and use the model from its run:

model = mlflow.pyfunc.load_model((f"runs:/{run_id}/model")) #return run_id from train.py
predictions = model.predict(new_data)

Load from Model Registry

If registered in the Model Registry:

model_uri = 'models:/dynamic_pricing_model/Production'
model = mlflow.pyfunc.load_model(model_uri)
predictions = model.predict(new_data)

---

Folder Structure

Dynamic_Pricing/
|-- artifacts/
|-- scripts/
    |-- __init__.py            
    |-- train.py            # Script to train and log the model
    |-- inference.py            # Script for single and batch inference
    |-- main.py            # Script to run the pipeline
    |-- preprocess.py            # Script to preprocess
|-- requirements.txt          # Dependencies for the project
|-- mlruns/                   # MLflow tracking directory
|-- data/                     # Sample input data
|-- README.md                 # Project documentation

---

MLflow Tracking and Management

To visualize and manage MLflow runs:

1. Start the MLflow UI:

   mlflow ui

2. Navigate to http://localhost:5000.
3. View logged runs, metrics, and artifacts.

---

Acknowledgements

• MLflow Documentation
• Scikit-learn Documentation

---

License

This project is licensed under the MIT License. See LICENSE for more details.

---

Future Enhancements

• Support for additional machine learning models.
• Integration with real-time pricing APIs.
• Enhanced data preprocessing pipelines.
• Deployment of the model for real-time inference.