17. GlossaryΒΆ
- accuracy evaluation
- Point-wise estimate of the magnitude of approximation error.
- array-like
- A Python iterable, NumPy array, pandas DataFrame or Series, which can represent a data sample (rectangular array) of the required dimension and data type. For example, 1D: a flat NumPy array, pandas Series, a list; 2D: a 2D NumPy array, a pandas DataFrame, a nested list; 3D: a 3D NumPy array or a multi-indexed pandas DataFrame. Note that pandas DataFrame and Series are currently supported only in GTApprox, GTDF, and GTDR methods.
- DoE
- Design of Experiment, the set of the input parts of training vectors.
- float32_epsilon
The machine single precision float 32-bit epsilon; you can get its numeric value as:
> import numpy > numpy.finfo(numpy.float32).eps
- float64_epsilon
The machine double precision float 64-bit epsilon; you can get its numeric value as:
> import numpy > numpy.finfo(numpy.float).eps
- internal validation
- A statistical procedure based estimating accuracy of the approximation technique on the given training data by Cross-validation procedure details - repeated construction of approximations using subsets of data followed by testing the approximations on the complementary subsets. See details in Internal Validation section.
- MAE
- mean average absolute error (see MAE)
- model gradient
- Gradient or, more generally, Jacobi matrix (if the output dimension is greater than 1) of the first order partial derivatives of the approximation model.
- Octave
- GNU Octave is a high-level interpreted language, primarily intended for numerical computations. It provides capabilities for the numerical solution of linear and nonlinear problems, and for performing other numerical experiments. Octave language is quite similar to Matlab so that most programs are easily portable.
- Partial correlation
- Correlation between pair of features with set of control features present. The influence of control features on considered is substracted and then correlation is computed between pair of remaining residuals.
- R^2
- coefficient of determination (see R^2)
- RMS
- root mean square error (see RMS)
- RRMS
- relative root mean square error (see RRMS)