Semi-automated preprocessing Python module for near infrared spectroscopic (NIRS) data.

nippy is a Python (3.6+) module for rapid exploration of different NIRS preprocessing methods. nippy collects and wraps the most common preprocessing methods and provides tools for quickly constructing preprocessing pipes with alternative preprocessing combinations. Aim of this module is to enable the user to quickly test multiple alternativ preprocessing techniques and test how that affects the performance of the NIRS model.

Comprehensive manual is still being worked on. For a simplified example of how nippy works you can look into the examples directory. We provide here a crash-course into how nippy can be used.

The typical structure of the nippy analysis is as follows:

1. Specify the methods you wish to try and the associated parameters by generating an INI-formatted configuration file. (for more detailed documentation about writing configuration files please check out the CONFIGURATION.md). For example, configuring nippy to test 2nd derivative Savitzky-Golay filtering (with 3rd order polynomial fit) at three different filter-lengths (7, 11 and 31 samples) can be accomplished by adding the following section to the configuration file.

[SAVGOL]
    filter_win = 7, 11, 31
    poly_order = 3
    deriv_order = 2
    also_skip = True

2. Load your NIR data into a numpy matrix (rows wavelengths, columns samples). Load your wavelengths into a numpy vector.

    data = np.genfromtxt('nir_data.csv', delimiter=',')
    wavelength = data[0, :]
    spectra = data[1:, :].T  # Rows = wavelength, Columns = samples

3. Import nippy and read your protocol file using nippy.read_configuration.

    import nippy
    pipelines = nippy.read_configuration('example_protocol.ini')

4. nippy generates a list of all possible preprocessing permutation. Pass your data and the list of pipelines to the nippy-function.

    datasets = nippy.nippy(wavelength, spectra, pipelines)

The variable datasets now contains a list of datasets that have been preprocessed according to the methods listed in the pipelines variable. Preprocessed data can be used in Python or exported for use in other applications.

numpy (1.13.1+)
scipy (0.19.1+)
sklearn (0.19.2+)

pip install git+https://github.com/uef-bbc/nippy

• nippy.py: contains all of the preprocessing operations
• handler.py: top-level script for generating and running multiple preprocessing pipelines

• example.py: example script for performing multiple preprocessing pipelines
• example.ini: example nippy protocol
• nir_data.csv: small NIR dataset for demonstration purposes