.. _tabular-solution:

================
Tabular Solution
================

The **Tabular** solution generates static artificial neural network (ANN) models for problems where both inputs and outputs are sets of scalar parameters (tabular data).

.. contents:: On this page
   :local:
   :depth: 2

Overview
========

NeurEco Tabular is designed for problems where:

- Your data is organized in tables (rows = samples, columns = features)
- Each sample is independent (no time dependency between samples)
- You want to learn the relationship between input features and output features

**Example:** Predicting a house's energy consumption based on temperature, wind speed, humidity, and other environmental factors.

How It Works
------------

You provide NeurEco with a dataset containing input features and corresponding output values. NeurEco automatically:

1. Starts from the smallest possible neural network
2. Enriches it step by step, adding neurons and connections
3. Saves intermediate models to checkpoints
4. Stops when optimal accuracy is reached

The result is a **parsimonious model** — significantly smaller than traditional deep learning approaches, with no tradeoff between size and accuracy.

Available Templates
===================

The Tabular solution offers three templates for different problem types:

Regression
----------

Use **Regression** when your outputs are **continuous numerical values**.

**Use cases:**

- Predicting physical quantities (temperature, pressure, energy)
- Estimating costs or prices
- Approximating simulation results

**Output format:** Continuous values in shape ``(n_samples, n_outputs)``

Classification
--------------

Use **Classification** when your outputs are **discrete categories or labels**.

**Use cases:**

- Categorizing items into classes
- Diagnostic classification (e.g., disease types)
- Quality control (pass/fail, grade levels)

**Output format:** One-hot encoded labels in shape ``(n_samples, n_classes)``

.. note::

   Classification outputs must be **one-hot encoded**: each row contains exactly one ``1`` indicating the class, with all other values being ``0``.

Compression
-----------

Use **Compression** when you want to **reduce data dimensionality** or create an autoencoder.

**Use cases:**

- Dimensionality reduction
- Data compression for IoT/embedded systems
- Feature extraction
- Anomaly detection (via reconstruction error)

**Output format:** No separate output files needed — the model learns to reconstruct the input.

**Unique feature:** Compression produces three models:

- **Complete model:** Full encoder + decoder
- **Compression model (encoder):** Input -> compressed representation
- **Decompression model (decoder):** Compressed representation -> reconstructed output

This is particularly useful for IoT applications where a small encoder runs on the device and a larger decoder runs on the server.

Choosing a Template
-------------------

.. list-table::
   :header-rows: 1
   :widths: 30 70

   * - Your Goal
     - Template
   * - Predict continuous values
     - Regression
   * - Classify into categories
     - Classification
   * - Compress/reconstruct data
     - Compression

Workflow Overview
=================

Regardless of which template you choose, the workflow follows these steps:

1. **Data Preparation** — Format your data correctly for NeurEco
2. **Build Model** — Train the model with your data
3. **Evaluate Model** — Test performance on unseen data
4. **Export Model** — Deploy in your target format

Each step can be performed using the **GUI**, **Python API**, or **CLI**.

.. toctree::
   :maxdepth: 2
   :caption: Tabular Documentation

   data-preparation
   building-models/index
   evaluating-models
   exporting-models
   advanced-features
   model-conversion
   examples/index