SmolAgent ML Assistant

This project is a Streamlit-based application that provides an end-to-end machine learning workflow. Users can upload a dataset, select features and a target variable, choose a task type (Regression or Classification), and run a predefined ML pipeline. The application uses a custom agent to preprocess data, train a model, evaluate it, and generate a reproducible Python script. Results include performance metrics, visualizations, and the generated code.

Features

• Upload a CSV dataset via a user-friendly Streamlit interface.
• Select target and feature columns for the ML task.
• Choose between Regression and Classification tasks with supported models (e.g., Linear Regression, Random Forest).
• Execute a full ML pipeline: preprocessing, training, and evaluation.
• Display performance metrics (e.g., MSE, R² for regression; accuracy, precision, recall for classification).
• Visualize results with Plotly (e.g., residual plots for regression, confusion matrices for classification).
• Generate a standalone Python script for the ML workflow.

Prerequisites

Before running the project, ensure you have the following installed:

• Python 3.11 (recommended for compatibility with Streamlit and dependencies).
• Docker: For containerization.
• Minikube: For local Kubernetes deployment.
• kubectl: To interact with Kubernetes.
• Git: To clone the repository.

Installation and Setup

1. Clone the Repository

git clone https://github.com/RobuRishabh/Smolagents.git
cd smolagent

2. Install Dependencies

Install the required Python packages:

pip install streamlit smolagents pandas numpy scikit-learn plotly pyyaml python-dotenv

Note: If smolagents is a custom package, ensure it's available in your repository or install it manually (e.g., pip install -e . if it's part of the project).

3. Configure Environment Variables

Create a .env file in the project root with the necessary environment variables:

# .env
API_BASE=http://127.0.0.1:11434
MODEL_ID=ollama_chat/qwen2.5-coder:3b

5. Run the Application Locally

streamlit run app.py

Open your browser at http://localhost:8501. Upload a CSV dataset, select your features, and run the agent.

---

Containerization with Docker

1. Create a Dockerfile

Create a file named Dockerfile in the project root with the following content:

# Use an official Python runtime as the base image
FROM python:3.11-slim

# Set the working directory
WORKDIR /app

# Copy the requirements file (create this file if not already present)
COPY requirements.txt .

# Install dependencies
RUN pip install --no-cache-dir -r requirements.txt

# Copy the project files
COPY . .

# Expose the Streamlit port
EXPOSE 8501

# Command to run the application
CMD ["streamlit", "run", "app.py", "--server.port=8501", "--server.address=0.0.0.0"]

2. Create a requirements.txt

Generate a requirements.txt file:

pip freeze > requirements.txt

Alternatively, manually create requirements.txt with the following:

streamlit==1.36.0
pandas
numpy
scikit-learn
plotly
pyyaml
python-dotenv

Note: Add smolagents if it's a PyPI package, or ensure its source code is included in the project.

3. Build the Docker Image

docker build -t smolagent-ml-assistant:latest .

4. Test the Docker Container Locally

docker run -p 8501:8501 --env-file .env smolagent-ml-assistant:latest

Access the app at http://localhost:8501.