Python 時間波形をnumpyのスライスを使ってダウンサンプリングする方法

import timeit

import numpy as np


def downsample(t: np.ndarray, data: np.ndarray, t_downsample: np.ndarray) -> np.ndarray:
    """
    Perform downsampling on time series data using slicing.

    Parameters
    ----------
    t : numpy.ndarray
        Original time axis.
    data : numpy.ndarray
        Original data corresponding to the time axis.
    t_downsample : numpy.ndarray
        New time axis for downsampling.

    Returns
    -------
    numpy.ndarray
        Downsampled data.
    """
    indices = ((t_downsample - t[0]) / (t[1] - t[0])).astype(int)
    extracted_data = data[indices]
    return extracted_data


def downsample_searchsorted(
    t: np.ndarray, data: np.ndarray, t_downsample: np.ndarray
) -> np.ndarray:
    """
    Perform downsampling on time series data using np.searchsorted.

    Parameters
    ----------
    t : numpy.ndarray
        Original time axis.
    data : numpy.ndarray
        Original data corresponding to the time axis.
    t_downsample : numpy.ndarray
        New time axis for downsampling.

    Returns
    -------
    numpy.ndarray
        Downsampled data.
    """
    closest_indices = np.searchsorted(t, t_downsample)
    extracted_data = data[closest_indices]
    return extracted_data


# Generate original data
t_original = np.arange(0, 100000, 1)
data_original = t_original * 10

# Specify the range and step size for the new time axis for downsampling
start_value = 0
end_value = 10000
step_size = 2
t_downsample = np.arange(start_value, end_value, step_size)

# Measure the time taken by each function
time_downsample = timeit.timeit(
    lambda: downsample(t_original, data_original, t_downsample),
    number=100,
)
time_searchsorted = timeit.timeit(
    lambda: downsample_searchsorted(t_original, data_original, t_downsample),
    number=100,
)

print(f"t_original       : {t_original}")
print(f"data_original    : {data_original}")
print(f"t_downsample     : {t_downsample}")
print(f"data_downsample  : {downsample(t_original, data_original, t_downsample)}")
# t_original       : [    0     1     2 ... 99997 99998 99999]
# data_original    : [     0     10     20 ... 999970 999980 999990]
# t_downsample     : [   0    2    4 ... 9994 9996 9998]
# data_downsample  : [    0    20    40 ... 99940 99960 99980]

# Output the results
print(
    f"Time taken by downsample function: {time_downsample:.6f} seconds"
)  # 0.003711 seconds
print(
    f"Time taken by downsample_searchsorted function: {time_searchsorted:.6f} seconds"
)  # 0.032072 seconds