eb-adapters 0.1.0

Adapter layer for third-party forecasting libraries in Electric Barometer

Electric Barometer Adapters (eb-adapters)

This repository contains the model adapter layer of the Electric Barometer ecosystem.

eb-adapters provides standardized interfaces that adapt common forecasting and machine-learning libraries to the Electric Barometer evaluation and readiness frameworks, enabling consistent metric computation and model comparison across heterogeneous modeling approaches.

Conceptual definitions and evaluation philosophy are maintained in the companion research repository: eb-papers.

---

Naming convention

Electric Barometer packages follow standard Python packaging conventions:

• Distribution names (used with pip install) use hyphens
  e.g. pip install eb-adapters
• Python import paths use underscores
  e.g. import eb_adapters

This distinction is intentional and consistent across the Electric Barometer ecosystem.

---

Role Within Electric Barometer

Within the Electric Barometer ecosystem:

• eb-papers defines concepts, frameworks, and meaning
• eb-metrics implements individual metrics
• eb-evaluation orchestrates evaluation workflows
• eb-adapters standardizes model interfaces

This repository focuses on bridging external modeling libraries into a common, evaluation-ready form.

---

What This Library Provides

• A common adapter base class defining a unified interface for forecast models
• Library-specific adapters for popular forecasting and ML frameworks
• Consistent prediction and evaluation hooks compatible with EB metrics
• Extensible patterns for adding new model families without changing downstream evaluation logic

Current adapters include support for:

• CatBoost
• LightGBM
• Prophet
• Statsmodels-based models

---

Scope

This repository focuses on model adaptation, not model training algorithms or metric definitions.

In scope:

• Wrapping external model APIs behind a consistent adapter interface
• Standardizing prediction outputs for evaluation
• Providing compatibility with EB evaluation pipelines

Out of scope:

• Metric definitions and loss formulations (see eb-metrics)
• Evaluation orchestration logic (see eb-evaluation)
• Model training methodology or hyperparameter optimization
• Conceptual framework definitions (see eb-papers)

---

Installation

Install from PyPI:

pip install eb-adapters

For development or local use:

pip install -e .

---

Package Structure

The repository follows a clean, modern Python package layout:

eb-adapters/
├── src/eb_adapters/
│   ├── base.py            # Abstract adapter base class
│   ├── catboost.py        # CatBoost model adapter
│   ├── lightgbm.py        # LightGBM model adapter
│   ├── prophet.py         # Prophet model adapter
│   └── statsmodels.py     # Statsmodels-based adapters
│
├── tests/
│   └── adapters/          # Unit tests for adapter implementations
│
├── pyproject.toml         # Build and dependency configuration
├── README.md              # Project documentation
└── LICENSE                # BSD-3-Clause license

---

Relationship to Other EB Repositories

• eb-papers
  Source of truth for conceptual definitions and evaluation philosophy.

• eb-metrics
  Provides the metric implementations used during evaluation.

• eb-evaluation
  Orchestrates evaluation workflows using adapted models.

• eb-adapters
  Ensures heterogeneous models can be evaluated consistently.

When discrepancies arise, conceptual intent in eb-papers should be treated as authoritative.

---

Development and Testing

Tests are located under the tests/ directory and mirror adapter coverage.

To run the test suite:

pytest

---

Status

This package is under active development. Public APIs may evolve prior to the first stable release.