Augmentation

Data augmentation is a mathematical transformation of the spectral data that adds stochastic artifacts that resemble the ones that can be found in real-world data. The objective of data augmentation is to increase the size of the dataset and to improve the generalization of the model. The following algorithms are available:

• Augmentation with normal noise
• Augmentation with uniform noise
• Augmentation with exponential noise
• Baseline shift
• Peak shift
• Scale spectrum

Augmentation with normal noise

Augmentation with normal noise to the spectrum. Gaussian noise with mean 0 and standard deviation defined by the user is added to each data-point of the spectrum.

Arguments

Usage Example

from chemotools.augmentation import NormalNoise

normal_noise = NormalNoise(scale=0.001)
augmented_spectra = normal_noise.fit_transform(spectra)

Plotting Example

Augmentation with uniform noise

Augmentation with uniform noise to the spectrum. Uniform noise with minimum and maximum values defined by the user is added to each data-point of the spectrum.

Arguments

Usage Example

from chemotools.augmentation import UniformNoise

uniform_noise = UniformNoise(min=-0.001, max=0.001)
augmented_spectra = uniform_noise.fit_transform(spectra)

Plotting Example

Augmentation with exponential noise

Augmentation of the spectra by adding noise following an exponential distribution with a given standard distribution.

Arguments

Usage Example

from chemotools.augmentation import ExponentialNoise

exponential_noise = ExponentialNoise(scale=0.001)
augmented_spectra = exponential_noise.fit_transform(spectra)

Plotting Example

Augmentation with baseline shift

Adds a baseline to the data. The baseline is drawn from a one-sided uniform distribution between 0 and 0 + scale.

Arguments

Usage Example

from chemotools.augmentation import BaselineShift

baseline = BaselineShift(scale=0.05)
augmented_spectra = baseline.fit_transform(spectra)

Plotting Example

Augmentation with peak shift noise

Augmentation of the spectra by shifting the peak a defined number of indices along the x-axis. This augmentation technique is specially interesting in Raman spectra, where peak shifts between or within instruments can occur as result of differences in the gratings.

Arguments

Usage Example

from chemotools.augmentation import IndexShift

shift = IndexShift(shift=2)
augmented_spectra = shift.fit_transform(spectra)

Plotting Example

Augmentation with peak scaling

Scales the data by a value drawn from the uniform distribution centered around 1.0 with

Arguments

Usage Example

from chemotools.augmentation import SpectrumScale

scale = SpectrumScale(scale=0.01)
augmented_spectra = scale.fit_transform(spectra)

Plotting Example