Using tsfel feature extraction settings

[Incrementis]

Greetings,

when testing the code in "https://github.com/predict-idlab/tsflex/blob/main/examples/tsfel_integration.ipynb", especially the code in "Using tsfel feature extraction settings (i.e., dictionaries)"

simple_feats = MultipleFeatureDescriptors(
    functions=tsfel_feature_dict_wrapper(get_features_by_domain("temporal")),
    series_names=["ACC_SMV", "EDA", "TMP"],
    windows=["5min", "2.5min"],
    strides=["2.5min"],
)
feature_collection = FeatureCollection(simple_feats)
feature_collection

This works for functions=tsfel_feature_dict_wrapper(get_features_by_domain("temporal")) and
functions=tsfel_feature_dict_wrapper(get_features_by_domain("statistical")), but not for
functions=tsfel_feature_dict_wrapper(get_features_by_domain("spectral")).
As this resolves into an error(see image below).

I understand that the relevant part of the error is this:
"AssertionError: Func FuncWrapper(lpcc, ['lpcc_1', 'lpcc_2', 'lpcc_3', 'lpcc_4', 'lpcc_5', 'lpcc_6', 'lpcc_7', 'lpcc_8', 'lpcc_9', 'lpcc_10', 'lpcc_11', 'lpcc_12', 'lpcc_13'], {'n_coeff': 12}) returned incorrect number of outputs (12)!"

To overcome the error the following code is used:

from tsfel.feature_extraction.features import (
    # spectral features
    fft_mean_coeff, fundamental_frequency, human_range_energy, lpcc,mfcc, 
    max_power_spectrum, max_frequency, median_frequency, power_bandwidth, 
    spectral_centroid, spectral_decrease, spectral_distance, spectral_entropy,
    spectral_kurtosis, spectral_positive_turning, spectral_roll_off, spectral_roll_on,
    spectral_skewness, spectral_slope, spectral_spread, spectral_variation,
    wavelet_abs_mean, wavelet_energy, wavelet_entropy, wavelet_std,
    wavelet_var  
)

# Tsfel Standard Parametereinstellungen
all_fs = 100
# Erstellen der Feature-Sammlung
fc = FeatureCollection(
    MultipleFeatureDescriptors(
        functions=[
        # Spectral
        FuncWrapper(fft_mean_coeff, fs=all_fs, nfreq=256, output_names=[f"FFT mean coefficient_{i}" for i in range(256)]),
        FuncWrapper(fundamental_frequency, fs=all_fs), 
        FuncWrapper(human_range_energy, fs=all_fs),
        FuncWrapper(lpcc, n_coeff=12, output_names=[f"LPCC_{i}" for i in range(12)]),
        FuncWrapper(mfcc, cep_lifter=22, fs=all_fs, nfft=512, nfilt=40,num_ceps=12, pre_emphasis=0.97, output_names=[f"MFCC_{i}" for i in range(12)]),
        FuncWrapper(max_power_spectrum, fs=all_fs),
        FuncWrapper(max_frequency, fs=all_fs),
        FuncWrapper(median_frequency, fs=all_fs),
        FuncWrapper(power_bandwidth, fs=all_fs),
        FuncWrapper(spectral_centroid, fs=all_fs), 
        FuncWrapper(spectral_decrease, fs=all_fs),
        FuncWrapper(spectral_entropy, fs=all_fs),
        FuncWrapper(spectral_distance, fs=all_fs),
        FuncWrapper(spectral_kurtosis, fs=all_fs),
        FuncWrapper(spectral_positive_turning, fs=all_fs),
        FuncWrapper(spectral_roll_off, fs=all_fs),
        FuncWrapper(spectral_roll_on, fs=all_fs),
        FuncWrapper(spectral_skewness, fs=all_fs),
        FuncWrapper(spectral_slope, fs=all_fs),
        FuncWrapper(spectral_spread, fs=all_fs), 
        FuncWrapper(spectral_variation, fs=all_fs),
        FuncWrapper(wavelet_abs_mean, function=scipy.signal.ricker, widths=np.arange(1,10), output_names=[f"Wavelet absolute mean_{i}" for i in range(0,9)]),
        FuncWrapper(wavelet_energy, function=scipy.signal.ricker, widths=np.arange(1,10), output_names=[f"Wavelet energy_{i}" for i in range(0,9)]),
        FuncWrapper(wavelet_entropy, function=scipy.signal.ricker, widths=np.arange(1,10), output_names=["Wavelet entropy"]),
        FuncWrapper(wavelet_std,function=scipy.signal.ricker, widths=np.arange(1,10), output_names=[f"Wavelet standard deviation_{i}" for i in range(0,9)]),
        FuncWrapper(wavelet_var,function=scipy.signal.ricker, widths=np.arange(1,10), output_names=[f"Wavelet variance_{i}" for i in range(0,9)]),
    ],
       #TODO: Set your specifically needed arguments
    )
)

The information for the arguments were filtered from the tsfel specific features.json file(use json_path = tsfel.__path__[0] + "/feature_extraction/features.json to find its location).

As you can imagine, the code takes quite a long time to write, so my question is:

• Maybe I overdid it here because maybe there is an easier way, but if so, is there an easier way and how?