Glossary

Definitions of terms used throughout the NeurEco documentation.

Build

The process of training a NeurEco model. During build, the algorithm automatically determines the optimal network architecture and parameters.

Checkpoint

A saved state during training that captures intermediate models. Checkpoints allow resuming training and selecting models with different accuracy/complexity trade-offs.

Classifier

A Tabular model that predicts discrete categories (classes) from input features. Outputs are class probabilities.

Complex output

An output value with both real and imaginary components. Used in PFS for frequency response data (magnitude and phase).

Compressor

A Tabular model that reduces data dimensionality. The encoder compresses data; the decoder reconstructs it.

Compute capability

NVIDIA’s measure of GPU features. NeurEco requires compute capability 7.5 or higher for GPU acceleration.

Convolutional Dynamic

A Dynamic model that applies a convolutional neural network to a sliding window over past input values to predict outputs. Requires uniform time steps. Called CNN Dynamic in the GUI; its Python class is ConvolutiveDynamic.

ConvolutiveDynamic

The Python class of the Convolutional Dynamic model (see Convolutional Dynamic).

Denominator size

In PFS models, the complexity of the rational function approximation. Larger denominator size captures more complex frequency behavior.

Dynamic solution

NeurEco solution for time-series data where past values influence future predictions. Includes ODE and Convolutional Dynamic approaches.

ECNN

File extension for Convolutional Dynamic models (.ecnn = embedded convolutional neural network).

EDNN

File extension for NeurEco Tabular models (.ednn = embedded deep neural network).

EFNN

File extension for PFS models (.efnn = embedded frequency neural network).

Embed license

Add-on license enabling export to deployment formats (ONNX, FMU, C, VBA).

Enrichment

In PFS, the iterative process of adding complexity to the rational approximation to improve accuracy.

ERNN

File extension for ODE models (.ernn = embedded recurrent neural network).

Evaluate

Running inference on a trained model to generate predictions for new input data.

Excitation

In Dynamic models, the input signal that drives the system. Also called input or stimulus.

Export

Converting a NeurEco model to a deployment format (ONNX, FMU, C header, VBA).

FMU

Functional Mock-up Unit. A standard format for simulation model exchange, widely used in automotive and industrial applications.

Gradient

The partial derivatives of model outputs with respect to inputs. Used for optimization and sensitivity analysis.

Hidden state

In ODE models, internal variables that capture system dynamics not directly observable in inputs/outputs.

Initialization

In Dynamic models, providing initial time steps to establish the system state before prediction begins.

L2 relative error

A measure of prediction accuracy: the L2 norm of prediction errors divided by the L2 norm of actual values. Lower is better.

Neuron

A computational unit in the neural network. NeurEco automatically determines the optimal number of neurons.

Normalization

Scaling data to a standard range. NeurEco supports various normalization strategies (mean shift, min-max scaling, etc.).

ODE

Ordinary Differential Equation. The ODE model learns hidden states that evolve according to differential equations.

ONNX

Open Neural Network Exchange. A portable format for machine learning models supported by many frameworks and runtimes.

Output

The predicted value(s) from a model. In Tabular, outputs are features; in Dynamic, outputs are time-series values.

PFS

Parametric Frequency Sweep. NeurEco solution for frequency-domain modeling with complex-valued outputs.

Rational function

A ratio of polynomials. PFS uses rational functions to approximate frequency response behavior.

Regressor

A Tabular model that predicts continuous numerical values from input features.

Sensitivity

How much a model output changes in response to input changes. Sensitivity analysis reveals which inputs most influence outputs.

Solution

A NeurEco problem type: Tabular, Dynamic, or PFS. Each solution addresses different data and modeling scenarios.

Steady state

In Dynamic models, the equilibrium condition where system outputs remain constant for constant inputs.

Tabular solution

NeurEco solution for structured data where each sample is independent. Includes Regressor, Classifier, and Compressor.

Template

A model type within a solution: Regression, Classification and Compression (Tabular), ODE and Convolutional Dynamic (Dynamic), and PFS. The CLI takes the template name first (e.g., neureco Regression build build.conf).

Trajectory

In Dynamic models, a complete time-series sequence from start to end. Training typically uses multiple trajectories.

Validation data

Data held out from training to monitor model performance and prevent overfitting.

VBA

Visual Basic for Applications. Export format for using NeurEco models in Microsoft Excel.

Window size

In Convolutional Dynamic models, the number of past time steps used to predict the current output (at least 3).

File Extensions Reference

Extension	Solution	Description
.ednn	Tabular	Regressor, Classifier, Compressor models
.ernn	Dynamic ODE	ODE models with hidden states
.ecnn	Dynamic Convolutional	Convolutional Dynamic sliding-window models
.efnn	PFS	Parametric Frequency Sweep models
.onnx	Tabular	ONNX export format
.fmu	All	FMU export format
.h	Tabular	C header export
.bas	Tabular	VBA export for Excel

Module Import Reference

Import	Purpose
from NeurEco import NeurEcoTabular as tab	Tabular solution (Regressor, Classifier, Compressor)
from NeurEco import NeurEcoDynamic as dyn	Dynamic solution (ODE, ConvolutiveDynamic)
from NeurEco import NeurEcoFrequential as freq	PFS solution
from NeurEco import NeurEcoError	Base exception class