Corerec v0.5.0

[vishesh9131]

Core Framework and Architecture

• Imshow Module: A new interactive visualization system with real-time plotting and customizable dashboards.
• Core Architecture Overhaul: A complete restructure with a new core/ directory, including a unified base_model.py, collision-free embedding tables, and a modular tower architecture.
• C++ Extensions: Native C++ extensions for high-speed tensor and embedding operations, with GPU acceleration support.
• Multimodal Support: A new framework for multimodal data fusion, handling text, image, and audio data.
• Hybrid Recommendation Systems: New modules for LLM-based reranking (prompt_reranker.py) and a two-stage retrieval-and-rerank pipeline.
• Retrieval Systems: A new framework for retrieval, including a dssm.py implementation for semantic search.
• Modular Tower Architecture: A new towers/ module with abstract and specific tower implementations (e.g., CNN, MLP, Transformer) for building complex models.
• Training Infrastructure: An advanced training system with callbacks, comprehensive metrics, online training capabilities, and distributed training support.
• Data Management: An enhanced data pipeline with new core data structures, multimodal dataset handling, and a streaming data loader.
• Optimal Path Module: A framework for path optimization and user journey planning using graph-based algorithms.
• Judge Module: A comprehensive model evaluation system for benchmarking, A/B testing, and model comparison.
• SSH Helper: Utilities for SSH connection management, remote execution, and secure file transfers.

---

Engine Improvements and Additions

• New Engine Implementations: Several new recommendation engines have been added, including Deep & Cross Network (dcn.py), Deep Factorization Machine (deepfm.py), Graph Neural Network Recommender (gnnrec.py), and Self-Attentive Sequential Recommendation (sasrec.py).
• Unionized Filter Engine Components: Added new components like Cornac BPR (cornac_bpr.py), Geometric Matrix Completion (geomlc.py), and Restricted Boltzmann Machine (rbm.py).
• Content Filter Engine Enhancements: Improved context personalization, graph-based algorithms, and embedding representation learning.
• Neural Network Base Components: New neural network implementations have been added, such as DCN, Caser, GRU, and NCF.
• Matrix Factorization Base Components: New components like ALS recommender and SVD recommender were introduced.
• Attention Mechanism Base Components: New attention-based algorithms like A2SVD and SASRec have been added.
• Bayesian Method Base Components: New components for Bayesian methods, including BPR and Multinomial VAE.
• Graph-Based Base Components: New graph-based algorithms like GeoIMC and LightGCN.

---

Other Additions

• Demo Frontends & Web Interface: A new demo_frontends/ module with multi-platform support (Netflix, Spotify, YouTube) and a RESTful backend API.
• Comprehensive Examples & Documentation: A wide range of new examples and quick-start guides for various engines and features.
• Comprehensive Test Suite: Extensive test coverage for new and existing components, including integration and smoke tests.
• Developer Tools: New utilities for configuration management, a hook system, structured logging, and reproducibility (seed.py).

---

This discussion was created from the release Corerec v0.5.0.

[vishesh9131]

see using connector method i have connected a pre cooked frontend :

from corerec.imshow import connector


def my_music_recommender(user_id, num_items=10):
    recommendations = []
    for i in range(num_items):
        recommendations.append({
                "id": f"song_{i}",
                "title": f"Great Song {i+1}",
                "artist": f"Artist {i+1}",
                "album": f"Album {i+1}"
        })
    return recommendations

connect = connector(
    predict_function=my_music_recommender,
    frontend="spotify",
    title="My Music Recommender",
    description="A simple music recommendation system.",
)

connect.run()