candidate choice in OLTW Dixon is offset by one

[sildater]

matchmaker/matchmaker/dp/oltw_dixon.py

Line 311 in d5c58d7

def select_candidate(self):

if min_idx <= len(norm_x_edge):
            self.candidate = np.array([self.ref_pointer - offset[0], min_idx])
        else:
            self.candidate = np.array(
                [min_idx - len(norm_x_edge), self.input_pointer - offset[1]]
            )

the pointer to the outer edge of the matrix in both matrices needs to be offset by one like so:

if min_idx < len(norm_x_edge):
            self.candidate = np.array([self.ref_pointer - offset[0] - 1, min_idx])
        else:
            self.candidate = np.array(
                [min_idx - len(norm_x_edge), self.input_pointer - offset[1] - 1]
            )

and the index comparison should probably be strict -> min_idx == len(norm_x_edge) means the minimzer is in the top right corner [0,-1] which is the second clause.

changing this index offset needs adjustments also in the assertion in select_next_direction

[sildater]

another place where the comparison probably should be strict is in select_next_direction: the conditional if self.input_pointer <= self.w: only if there is strictly less input than the window the next direction should be forced in this input, right?