octolearn 0.10.3

Structured AutoML Pipeline with Intelligent Dataset Profiling

OctoLearn Architecture Guide

Welcome to the complete architectural reference for OctoLearn — an enterprise-grade AutoML library built for transparency, robustness, and ease of use. This document explains how the library is built, why specific design choices were made, and how to extend it.

---

1. System Overview

OctoLearn follows a Pipeline Orchestration pattern. The central AutoML class acts as the conductor, coordinating specialized workers (Profiler, Cleaner, Trainer, ReportGenerator, etc.) to transform raw data into a production-ready model and a comprehensive PDF report.

High-Level Data Flow

graph TD
    A[Data Ingestion] --> B{Profiling & Health};
    B -->|High Risk| C[Risk Scoring & Alerts];
    B -->|Normal| D[Preprocessing Orchestrator];
    C --> D;
    
    subgraph "Preprocessing Orchestrator"
    D --> D1[ID & Constant Drops];
    D1 --> D2[Missing Value Imputation];
    D2 --> D3[Categorical Encoding];
    D3 --> D4[Feature Engineering];
    D4 --> D5[Scaling & Normalization];
    end
    
    D5 --> E[Model Arena];
    
    subgraph "Model Arena"
    E --> E1[XGBoost];
    E --> E2[LightGBM];
    E --> E3[Random Forest];
    E --> E4[Logistic Regression];
    E --> E5[Stacking Ensemble];
    end
    
    E1 & E2 & E3 & E4 & E5 --> F[Model Evaluation];
    F --> G[Comparison & Ranking];
    G --> H[Artifact Generation];
    
    subgraph "Reporting"
    H --> H1[PDF Intelligence Report];
    H --> H2[Model Registry Storage];
    H --> H3[SHAP Interpretation];
    end

Why a Pipeline Orchestrator?

The orchestrator pattern keeps each component single-responsibility and independently testable. You can swap out the cleaner, trainer, or report generator without touching the others. It also makes the execution order explicit and auditable.

---

2. Directory Structure

OctoLearn/
├── octolearn/
│   ├── core.py                    # AutoML orchestrator + config dataclasses
│   ├── config.py                  # Global constants (Optuna, model registry)
│   ├── profiling/
│   │   └── data_profiler.py       # Statistical analysis → DatasetProfile
│   ├── preprocessing/
│   │   └── auto_cleaner.py        # Imputation, encoding, scaling
│   ├── models/
│   │   ├── model_trainer.py       # Multi-model training + Optuna
│   │   └── registry.py            # Model versioning and persistence
│   ├── experiments/
│   │   ├── report_generator.py    # PDF report (ReportLab)
│   │   ├── plot_generator.py      # matplotlib/seaborn visualizations
│   │   └── recommendation_engine.py # Narrative Summary Engine
│   ├── evaluation/
│   │   └── metrics.py             # Scoring functions
│   ├── utils/
│   │   └── helpers.py             # Logging, utilities
│   ├── fonts/                     # ShantellSans TTF font files
│   └── images/                    # logo.png
├── tests/
│   └── test_complete_pipeline.py  # Integration test suite
├── ARCHITECTURE.md                # This file
├── README.md                      # High-level overview
├── guide.md                       # User manual & cookbook
└── testing.md                     # Benchmarking & QA protocols

---

!!! tip "NumPy Docstring Standard" Starting from version 0.10.3, OctoLearn has adopted the NumPy Docstring Standard across the entire codebase. Every public class and method is documented with:

*   **Parameters**: Detailed type information and descriptions.
*   **Returns**: Clear explanation of output types and semantics.
*   **Attributes**: Internal state documentation for class instances.
*   **Examples**: Doctype-runnable examples for quick onboarding.

---

4. Configuration System (core.py)

Design: Dataclasses over kwargs

OctoLearn uses Python @dataclass objects instead of a flat list of keyword arguments. This provides:

• Type safety: IDE autocomplete and type checkers work correctly
• Grouping: Related settings are co-located (e.g., all Optuna settings in OptimizationConfig)
• Defaults: Each field has a sensible default, so AutoML() works out of the box
• Discoverability: Users can explore configs with help(OptimizationConfig)

Config Objects

Class	Key Fields	Rationale
DataConfig	sample_size=5000, test_size=0.2, stratify_target=True	Sampling prevents OOM on large datasets
ProfilingConfig	detect_outliers=True, analyze_interactions=True	Both are expensive; can be disabled for speed
PreprocessingConfig	imputer_strategy, scaler='standard', encoder_strategy	Sensible defaults; user can override per-column
ModelingConfig	n_models=5, models_to_train=None, evaluation_metric=None	Metric auto-detected from task type
OptimizationConfig	optuna_trials_per_model=20, optuna_timeout_seconds=300	Bayesian optimization speed/quality tradeoff
ReportingConfig	report_detail='detailed', visuals_limit=10	"Dashboard" vs "Simple" plot modes
ParallelConfig	n_jobs=1, backend='loky'	Sequential Optuna for Windows safety

The fit() Override Pattern

fit() accepts optional keyword arguments that temporarily override config values for a single run. Internally, the original config values are snapshotted, overrides applied, pipeline executed, then originals restored in a finally block — ensuring non-destructive experimentation.

---

5. Intelligence Reporting (experiments/)

The reporting engine has evolved from static charts to Contextual Intelligence.

Contextual Mathematical Narrative

Rather than simply outputting raw statistics, OctoLearn generates a Mathematical Narrative. It automatically interprets Pearson correlation coefficients, evaluates feature variance, and explains multi-collinearity in plain English. This guides stakeholders through a transparent "Data Journey," translating abstract metrics into concrete business impact.

High-Dimensionality Fallbacks

When a dataset exceeds 10 features, the report automatically switches from a full correlation matrix to a Top-N Correlation Bar Chart. This prevents "visual noise" and focuses the analyst's attention on the most impactful predictive signals.

Model Arena

The ModelTrainer reports results via the "Model Arena" — a competitive leaderboard that ranks models across multiple dimensions (Accuracy, F1, Latency, Calibration). The best model is promoted to the "Champion" slot, but full benchmarks are preserved for auditability.

---

6. Data Profiling (profiling/data_profiler.py)

Output: DatasetProfile

A comprehensive metadata container storing:

• Semantic Type Inference: Distinguishes between 'numeric', 'categorical', 'id', 'date', and 'text' columns using layered heuristics.
• Deep Statistical Boundaries: Computes exhaustive descriptive statistics (Mean, Median, Std, Skewness, Kurtosis) for every numeric feature, flagging severe distribution skew (>2.0) out-of-the-box.
• Data Quality Scoring: A 0-100 score based on missingness, class imbalance, duplication, and potential leakage suspects.
• Constraint Detection: Identifies constant columns and high-cardinality/low-variance features early to save downstream training time.

---

7. Data Cleaning (preprocessing/auto_cleaner.py)

!!! info "The Leakage Prevention Rule" fit_transform on Train only. transform on Test. This prevents statistical leakage (e.g., test data means influencing train data imputations).

Industrial-Strength Pipeline

1. Adaptive Imputation: Uses mode for categories and mean/median for numerics based on distribution skew.
2. Rare Category Grouping: Prevents feature explosion and overfitting by grouping low-frequency labels.
3. Cardinality Management: Smart selection between One-Hot and Ordinal encoding based on feature cardinality and task type.

---

8. Model Training (models/model_trainer.py)

Bayesian Optimization (Optuna)

Optuna uses Tree-structured Parzen Estimators (TPE) to intelligently search the hyperparameter space. This approach builds a probabilistic model of "good" parameter regions, finding optimal settings significantly faster than Grid or Random Search.

Stacking Ensembles

For production-grade performance, ModelTrainer can generate a Stacking Ensemble. This uses the top-performing base models (e.g., XGBoost, LightGBM) as "voters" and a meta-model (Logistic Regression) to compute the final prediction, often yielding a 1-3% boost in primary metrics.

---

9. Model Registry (models/registry.py)

OctoLearn includes a local Model Registry for version control.

• Versioned Artifacts: Models are saved as .pkl files with version stamps.
• Metadata Database: A JSON backend tracks performance metrics, training timestamps, and hyperparameters for every version.
• Safe Serialization: Handles complex NumPy/Pandas return types safely during JSON serialization.

---

10. How to Build & Verify

Running the Industry Benchmark

python test_complete_pipeline.py

This script exercises all 6 phases of the pipeline across diverse datasets to ensure 0-regression performance.

---

11. Key Architecture Principles

Principle	Implementation
Controllability	Fine-grained configuration via nested dataclasses.
Reproducibility	Global random state management in DataConfig.
Observability	Real-time logging and detailed PDF intelligence reporting.
Portability	Pure Python dependency stack (no external DBs or binaries).

---

OctoLearn Architecture v0.10.3 — Updated 2026-02-21