Fix calculation of direction coverage

[stklik]

It seems that the pipeline uses the following direction_coverage function for the road analysis.

I found two issues:

• first n_bins is 25 by default, but should be 35, so it's 36 bins in total. The calculated data is thus split differently than the comment indicates
• second, the code uses np.arccos(dot_product) for the calculation of the angle. However, np.arccos produces values in the range [0, π] (i.e. [0, 180] degrees). In other words, this function calculates the smallest angle between THE_NORTH and the segment, independent of the side.

I suggest the following code to fix it (note, it uses numpy throughout and is thus faster than the previous method).

def get_thetas(xs, ys, skip=1):
    """Transform x,y coordinates of points and return each segment's offset from x-axis in the range [np.pi, np.pi]"""
    xdiffs = xs[1:] - xs[0:-1]
    ydiffs = ys[1:] - ys[0:-1]
    thetas = np.arctan2(ydiffs, xdiffs) 
    return thetas

def dir_cov(the_test, n_bins=36):
    np_arr = np.array(the_test)
    thetas, kappas = get_thetas(np_arr[:,0], np_arr[:,1])
    coverage_buckets = np.linspace(-np.pi, np.pi, num=n_bins)
    covered_elements = set(np.digitize(thetas, coverage_buckets))
    return len(covered_elements) / len(coverage_buckets)

According to my tests on the data, this behaviour should be correct.