gradia 1.0.0

Local-first ML training visualization and tracking dashboard.

G R A D I A

Next-Generation Local-First MLOps Platform

---

🚀 Overview

Gradia is a high-performance, asynchronous monitoring solution designed for local machine learning workflows. Unlike cloud-native behemoths, Gradia focuses on zero-latency, privacy-first tracking that runs directly alongside your training loop.

Built on FastAPI, WebSockets (simulated via high-frequency polling), and Reactive UI, Gradia provides granular visibility into your model's training dynamics, system resource consumption, and feature importance without the overhead of external servers.

⚡ Key Capabilities

Feature	Description
Real-Time Telemetry	Nanosecond-precision tracking of Loss, Accuracy, and custom metrics via async event dispatching.
Intelligent Auto-Discovery	Heuristic analysis of tabular datasets to automatically infer task types (Regression vs Classification) and suggest optimal architectures (CNNs, RFCs).
System Profiling	Integrated psutil hooks for monitoring CPU/GPU* and RAM saturation during training epochs.
Artifact Management	Automated checkpointing (best-ckpt) and structured logging (events.jsonl) with thread-safe IO.
Comprehensive Reporting	One-click generation of audit-ready PDF/JSON reports containing full training history and confusion matrices.
Interactive Evaluation	Post-training validation suite featuring dynamic Heatmap visualization for Confusion Matrices.

🛠️ Architecture

Gradia employs a Producer-Consumer architecture:

1. Trainer Thread (Producer): Executes the Scikit-Learn training loop, emitting atomic events to a thread-locked EventLogger.
2. System Thread: Asynchronously samples hardware metrics.
3. Visualization Server (Consumer): A lightweight FastAPI instance that aggregates logs and serves a reactive Single Page Application (SPA).

This decoupling ensures that monitoring never bottlenecks your training throughput.

📦 Installation

pip install gradia --upgrade

💻 Usage

Quick Start

Initialize the environment and start the observer in one command. Gradia will auto-detect any CSV files in the directory.

gradia run .

Advanced CLI

Override default heuristics and bind to specific interfaces.

gradia run . \
  --target "churn_label" \
  --port 8080 \
  --workers 4

📊 Dashboard

Access the dashboard at http://localhost:8000.

• Configure: Select your model (Random Forest, MLP, etc.) and hyperparameters.
• Observe: Watch metrics stream in real-time.
• Analyze: Use the built-in Feature Importance charts to debug model bias.

🤝 Contributing

We welcome contributions! Please see CONTRIBUTING.md for details on submitting logical PRs.

📄 License

Distributed under the MIT License. See LICENSE for more information.

---

_{Built with ❤️ by STiFLeR for the ML Community.}