Add TranslationSuggester and friends

[ddaspit]

• replace Mockito with Decoy for mocking

---

This change is 

[codecov-commenter]

Codecov Report

Attention: 65 lines in your changes are missing coverage. Please review.

Files	Coverage Δ
machine/jobs/nmt_engine_build_job.py	72.60% <100.00%> (ø)
machine/tokenization/__init__.py	100.00% <100.00%> (ø)
machine/translation/__init__.py	100.00% <100.00%> (ø)
machine/translation/ecm_score_info.py	100.00% <100.00%> (ø)
...hine/translation/interactive_translation_engine.py	76.92% <100.00%> (ø)
...chine/translation/interactive_translation_model.py	75.00% <100.00%> (ø)
machine/translation/translation_constants.py	100.00% <100.00%> (ø)
machine/translation/translation_result_builder.py	90.19% <100.00%> (+51.63%)	⬆️
machine/translation/translation_suggestion.py	100.00% <100.00%> (ø)
machine/translation/word_edit_distance.py	100.00% <100.00%> (ø)
... and 17 more

... and 2 files with indirect coverage changes

📢 Thoughts on this report? Let us know!.

[johnml1135]

machine/translation/edit_distance.py line 57 at r1 (raw file):

                dist_matrix[i][j], _, _, _ = self._process_dist_matrix_cell(
                    x, y, dist_matrix, use_prefix_del_op, j != y_count or is_last_item_complete, i, j
                )

Is there a way to speed this up more? It looks very slow - list comprehension? cython? I am assuming that this will be computed many times...

[johnml1135]

machine/translation/edit_distance.py line 63 at r1 (raw file):

    def _process_dist_matrix_cell(
        self, x: Seq, y: Seq, dist_matrix: List[List[float]], use_prefix_del_op: bool, is_complete: bool, i: int, j: int
    ) -> Tuple[float, int, int, EditOperation]:

Do we need to try to convert this to C or cython code?

[ddaspit]

Reviewable status: 0 of 29 files reviewed, 2 unresolved discussions (waiting on @johnml1135)

---

machine/translation/edit_distance.py line 57 at r1 (raw file):

Previously, johnml1135 (John Lambert) wrote…

Is there a way to speed this up more? It looks very slow - list comprehension? cython? I am assuming that this will be computed many times...

This is a dynamic programming algorithm. Generally, the best computational complexity that you can achieve for this algorithm is O(m*n). You can make small optimizations, but not anything significant.

---

machine/translation/edit_distance.py line 63 at r1 (raw file):

Previously, johnml1135 (John Lambert) wrote…

Do we need to try to convert this to C or cython code?

This method is fairly simple. I don't think that converting it to C or Cython will give us much of a boost in performance.

[ddaspit]

I'm going to go ahead and merge this in, so I can start using it in silnlp. We can defer any optimizations until we have known performance issues.