User Guide¶

This guide explains the Foundation PLR pipeline architecture and how to use each component.

Pipeline Overview¶

The Foundation PLR pipeline consists of four main stages:

graph TB
    subgraph "Stage 1: Outlier Detection"
        OD[11 Methods]
        OD1[pupil-gt<br/>Ground Truth]
        OD2[MOMENT/UniTS/TimesNet<br/>Foundation Models]
        OD3[LOF/SVM/PROPHET<br/>Traditional]
    end

    subgraph "Stage 2: Imputation"
        IMP[8 Methods]
        IMP1[pupil-gt<br/>Ground Truth]
        IMP2[SAITS/CSDI<br/>Deep Learning]
        IMP3[MOMENT<br/>Foundation Model]
    end

    subgraph "Stage 3: Featurization"
        FEAT[Handcrafted Features]
        FEAT1[Amplitude Bins]
        FEAT2[Latency Features]
    end

    subgraph "Stage 4: Classification"
        CLS[CatBoost<br/>Fixed Classifier]
    end

    OD --> IMP --> FEAT --> CLS

Sections¶

Pipeline Stages¶

• Pipeline Overview - Detailed architecture explanation
• Outlier Detection - Available methods and configuration
• Imputation - Signal reconstruction approaches
• Featurization - Feature extraction from PLR signals
• Classification - Model training and evaluation

Infrastructure¶

• Prefect Orchestration - Workflow orchestration with Prefect