Moving linear regression indicator with numba

[andreas-vester]

I am trying to create a moving linear regression indicator and I wanted to utilize numba. However, I am struggling with the latter part.

Here's what I have so far. It is working, however, without applying numba it is quite slow once you throw large arrays at it.

import numpy as np

def ols_1d(y, window):
    y_roll = np.lib.stride_tricks.sliding_window_view(y, window_shape=window)

    m = list()
    c = list()
    for row in np.arange(y_roll.shape[0]):
        A = np.vstack([np.arange(1, window + 1), np.ones(window)]).T
        tmp_m, tmp_c = np.linalg.lstsq(A, y_roll[row], rcond=None)[0]
        m.append(tmp_m)
        c.append(tmp_c)

    m, c = np.array([m, c])

    return np.hstack((np.full((window - 1), np.nan), m * window + c))


def ols_2d(y, window):
    out = list()
    for col in range(y.shape[1]):
        out.append(ols_1d(y=y[:, col], window=window))

    return np.array(out).T


if __name__ == "__main__":
    a = np.random.randn(10000, 10)  # function is slow once you really increse number of columns

    print(ols_2d(a, 10))

The indicator is actually computing a linear regression applying np.linalg.lstsq function (https://numpy.org/doc/stable/reference/generated/numpy.linalg.lstsq.html) for a given window length. It than basically outputs the very last point of the regression line and moves on to the next range and computes the linear regression again. At the end of the day, ols_1d outputs the last point of every individual regression line putting it into an array.

Now, I need help in order to apply numba on top of it. I am not familiar with numba, but as far as my own trial and error goes, there might be an issue utilizing nb.lib.stride_tricks.sliding_window_view().

[polakowo]

@andreas-vester sorry, don't have time to help with that, better ask on Numba forum: https://numba.discourse.group