Rossmann Sales Forecast with GCP Deployment

Project Description

Rossmann, a chain of over 3,000 drug stores across 7 European countries, faces the challenge of predicting daily sales up to six weeks in advance.

---

Objective

This project focuses on forecasting store sales, taking into account various influencing factors, such as promotions, competition, holidays, seasonality, and locality.

---

Tech Stack

• Language: Python
• Cloud: Google Cloud Platform (GCP)
• UI: Flask
• Libraries: scikit-learn, pandas, numpy
• Code Management: Git, Github

---

Model Deployment

We deploy the model created using Google Cloud Platform (GCP). We explore two deployment approaches:

1. Traditional Approach
2. Dockerized Approach

Traditional Approach

In the traditional approach, we rent a cloud server, set up the necessary environment, deploy the model interface (Flask/Streamlit), and expose the required components for model deployment.

Dockerized Approach

The Dockerized approach involves packaging the model code and configurations into a Docker image, simplifying deployment and ensuring consistency across environments.

---

Deployment Steps

To deploy the model using GCP, follow these steps:

1. Create a source repository in GCP.

2. Set up a cloud build trigger for the GCP repository.

3. Create a Virtual Machine (VM) in GCP under "VM Instances."

4. Clone the repository in the VM.

5. Install Docker on the VM.

6. Choose one of the following deployment methods:

   • Setting up a Server Inside the VM:

     • Execute the provided commands to create and set up the server environment within the VM.

   • Setting up a Server Using Docker:

     • Pull the Docker image from GCP Container Registry.
     • Run the Docker image to deploy the model.

The necessary shell scripts are included for convenience:

• install-docker.sh: Installs Docker on the VM.
• setup-new-vm.sh: Sets up a new server inside the VM or deploys the model using Docker.

---