README for Predictive Modeling in Sequential Recommendation: Bridging Performance Laws with Data Quality Insights

---

PerformanceLaw Library

This project provides the PerformanceLaw Python library for evaluating the complexity and information quality of sequence data. The library includes several metrics:

Main Functions

• actual_entropy(seq): Estimates actual entropy of a sequence.
• actual_entropy_tq(seq): Same as above, shows a progress bar (needs tqdm).
• ApEn(U, m, r): Calculates the Approximate Entropy of a sequence U, with embedding dimension m and threshold r.
• compression_ratio(data): Measures the compressibility of a list of integers. Low ratio means data is more compressible (less random).
• shannon_entropy(sequence): Computes the Shannon entropy (in bits) of a sequence.

Installation

From the project root (where setup.py is located), run:

pip install -e .

Make sure you have installed numpy (and optionally tqdm):

pip install numpy tqdm

Example Usage

from PerformanceLaw import (
    actual_entropy,
    actual_entropy_tq,
    ApEn,
    compression_ratio,
    shannon_entropy
)

seq = [1, 2, 1, 2, 3]

print(actual_entropy(seq))
print(shannon_entropy(seq))
print(ApEn(seq, m=2, r=0.2))
print(compression_ratio(seq))

---

Application Scenarios

These functions help measure sequence complexity, data randomness, and information content. They can be used in recommendation systems, time-series analysis, and data quality evaluation.

---

For details, see source code comments and examples.

This README provides an overview of the project and instructions on how to navigate and utilize the different components available in the /General_Transformer and /Performance_Law_Appendix_Result directories.

Project Overview

The project focuses on advancing sequential recommendation systems through innovative models and performance law fitting strategies. It is divided into two main components:

1. General Transformer for Sequential Recommendation: Located in the /General_Transformer directory, this component implements a general transformer architecture for sequential recommendation tasks.
2. Performance Law Fitting Analysis: Found in the /Performance_Law_Appendix_Result directory, this component focuses on fitting performance laws for metrics like HR (Hit Rate) and NDCG (Normalized Discounted Cumulative Gain).

Directory Structure

• /General_Transformer: Contains scripts and code for training and evaluating transformer models for recommendation systems.
• /Performance_Law_Appendix_Result: Includes scripts and generated results for performance law fitting analysis, along with supplementary images referenced in the research paper.

Details

General Transformer for Sequential Recommendation

The scripts in the /General_Transformer directory are designed to train transformer models tailored for sequential recommendation tasks. The main features are:

• Model Training: Utilizing DDP (Distributed Data Parallel) to efficiently train on multiple GPUs.
• Hyperparameter Configurations: Flexible adjustments for layers, heads, batch sizes, and more.
• Logging and Evaluation: Detailed performance metrics are logged using libraries like WandB.

To learn more about using these scripts, refer to the README provided within the /General_Transformer directory.

Performance Law Fitting Analysis

Located in the /Performance_Law_Appendix_Result, this segment of the project analyzes performance laws through an innovative fitting approach. Key elements include:

• Performance Law Fitting: Detailed scripts for fitting performance laws to key metrics.
• Supplementary Images: Includes images such as PerformanceLaw_HR, PerformanceLaw_NDCG, ScalingLaw_HR, and ScalingLaw_NDCG for deeper insights and validation of research findings these images serve as supplementary material for the paper.

To understand the scripts and their execution, refer to the README within the /Performance_Law_Appendix_Result directory.

Getting Started

1. Navigate to the relevant directory:

   • For transformer models, explore /General_Transformer.
   • For performance law analysis, visit /Performance_Law_Appendix_Result.

2. Install required dependencies: Ensure all necessary Python libraries are installed as indicated in the README files within each directory.

3. Run the scripts: Follow the instructions to execute model training, evaluation, or performance fitting as required.

4. Explore Results and Graphs: Analyze outputs, performance metrics, and graphical results included in each component.