Using built-in metrics

In this tutorial, we will see how to access and use the built-in metrics available in RamanSPy.

import numpy as np

import ramanspy

To access the built-in metrics, simply use the ramanspy.metrics module.

Before we make use of the metrics, let us first define some dummy data to work with. We will create two spectra, one that is a sine wave and another that is a cosine wave. We will then use the metrics to compare the two spectra.

spectral_axis = np.linspace(100, 3600, 1500)

sine_data = np.sin(spectral_axis/120)
cosine_data = np.cos(spectral_axis/120)

sine_spectrum = ramanspy.Spectrum(sine_data, spectral_axis)
cosine_spectrum = ramanspy.Spectrum(cosine_data, spectral_axis)

We can visualise the two spectra with the data visualisation tools in RamanSPy.

_ = ramanspy.plot.spectra([sine_spectrum, cosine_spectrum], plot_type="single", label=["Sine", "Cosine"])

We can then measure the distance or similarity of the two spectra using the built-in metrics. For instance, we can use the ramanspy.metrics.MAE() to measure the mean absolute error between the two spectra.

ramanspy.metrics.MAE(sine_spectrum, cosine_spectrum)

0.892424

To double-check that the metric is working as expected, we can also measure the distance between the seme spectrum.

ramanspy.metrics.MAE(sine_spectrum, sine_spectrum)

0.0

Other metrics that are available include the MSE(), RMSE(), SAD(), SID(). For more information about these metrics, refer to their documentation.

For instance, we can use the SAD() to measure the spectral angle distance between the two spectra.

ramanspy.metrics.SAD(sine_spectrum, cosine_spectrum)

1.5558872986720789