Metalearning Library (metalearning_class)

A Python client library for meta-learning, designed to interface with the Panaceia/Sofon Meta-Learning Platform.

This library focuses on developing advanced machine learning techniques that "learn how to learn." It provides a robust, extensible framework for building meta-learning models leveraging neural networks and transformer architectures to solve complex regression and classification challenges efficiently.

🚀 Features

• Meta-Learning Development: Facilitates the creation and rapid experimentation of meta-learning algorithms.
• Key Components: Provides implementations for essential meta-learning components such as meta-learners and task encoders.
• Flexible Architectures: Supports diverse applications, from standard deep neural networks to custom Transformers and Robust RNNs.
• Platform Integration: Automatically synchronizes local model architectures, weights, and telemetry with the Sofon backend.

📦 Installation

To install the library, run the following command (requires Python 3.8+):

pip install metalearning_class

🛠️ Quick Start

Below is a basic workflow demonstrating how to initialize the client, authenticate, load data, and start working on a Sofon challenge.

[!TIP] Security Best Practice: Never hardcode your credentials. Use a .env file and python-dotenv to securely load your Sofon username and password.

import os
import metalearning_class as mtl
from dotenv import load_dotenv

# Load environment variables (.env)
load_dotenv()

# 1. Initialize the metalearning client
# Set gpu=True if you have a CUDA-compatible GPU configured
ml = mtl.Metalearning(gpu=False)

# 2. Login securely with your Sofon account
ml.login(os.getenv("SOFON_USERNAME"), os.getenv("SOFON_PASSWORD"))

# 3. Subscribe to a challenge and download the dataset
train_data = ml.subscribe_and_get_task("challenge_taskname")

# 4. Build and train your topology
# (Implementation depends on the specific challenge and architecture)
# ...

⚠️ Important Considerations

• Authentication Dependency: The library requires active communication with the Sofon backend. Most data retrieval and submission workflows will fail if you skip the ml.login() step.
• Data Types: Always cast your Pandas DataFrame features to numpy.float32 arrays before passing them to the training or inference pipelines to ensure compatibility with underlying TensorFlow operations.
• Mandatory Tracking: Ensure required tracking methods (like subscribe_and_get_task and setting topology names) are executed before attempting to define models or run predictions to avoid missing metadata errors.

📝 Recent Updates

v0.1.9 (2026-06-04)

• Added native support for base64-encoded and gzip-compressed backend training data payloads. Directly converts to Pandas DataFrame to optimize performance and reduce memory consumption.

v0.1.8 (2026-05-19)

• Standardized load_test_input preprocessing and input formatting with the training data pipeline.

(For a full history of changes, see the CHANGELOG.md file.)

🤝 Contributing

We will be transitioning to open-source soon and look forward to welcoming contributions to the Metalearning Library!

© 2026 Panaceia – All Rights Reserved