Reversed peptide aa scores hotfix (#417)

* reverse aa scores hotfix

* reverse aa scores hotfix

casanovo/denovo/model.py

@@ -684,7 +684,7 @@ def _get_top_peptide(
                 yield [
                     (
                         pep_score,
-                        aa_scores,
+                        aa_scores[::-1] if self.decoder.reverse else aa_scores,
                         "".join(self.decoder.detokenize(pred_tokens)),
                     )
                     for pep_score, _, aa_scores, pred_tokens in heapq.nlargest(

tests/unit_tests/test_unit.py

@@ -240,6 +240,25 @@ def test_beam_search_decode():
         [pep[-1] for pep in list(model._get_top_peptide(test_cache))[0]]
     ) == {"PEPK", "PEPP"}
 
+    # Test reverse aa scores when decoder is reversed
+    pred_cache = {
+        0: [(1.0, 0.42, np.array([1.0, 0.0]), torch.Tensor([4, 14]))]
+    }
+
+    model.decoder.reverse = True
+    top_peptides = list(model._get_top_peptide(pred_cache))
+    assert len(top_peptides) == 1
+    assert len(top_peptides[0]) == 1
+    assert np.allclose(top_peptides[0][0][1], np.array([0.0, 1.0]))
+    assert top_peptides[0][0][2] == "EP"
+
+    model.decoder.reverse = False
+    top_peptides = list(model._get_top_peptide(pred_cache))
+    assert len(top_peptides) == 1
+    assert len(top_peptides[0]) == 1
+    assert np.allclose(top_peptides[0][0][1], np.array([1.0, 0.0]))
+    assert top_peptides[0][0][2] == "PE"
+
     # Test _get_topk_beams().
     # Set scores to proceed generating the unfinished beam.
     step = 4