fixed bad prediction bug

casanovo/denovo/model.py

@@ -948,17 +948,23 @@ class DBSpec2Pep(Spec2Pep):
     Hijacks teacher-forcing implemented in Spec2Pep and uses it to predict scores between a spectra and associated peptide
     """
 
-    num_pairs = 1024
+    num_pairs = 2
 
     def __init__(self, *args, **kwargs):
         super().__init__(*args, **kwargs)
 
     def predict_step(self, batch, *args):
         batch_res = []
-        for indexes, t_or_d, new_batch in self.smart_batch_gen(batch):
+        for (
+            indexes,
+            t_or_d,
+            peptides,
+            precursors,
+            encoded_ms,
+        ) in self.smart_batch_gen(batch):
             with torch.set_grad_enabled(True):  # Fixes NaN!?
-                encoded_ms, precursors, peptides = new_batch
                 pred, truth = self.decoder(peptides, precursors, *encoded_ms)
+                print(truth)
                 sm = torch.nn.Softmax(dim=2)  # dim=2 is very important!
                 pred = sm(pred)
                 score_result, per_aa_score = calc_match_score(
@@ -994,7 +1000,7 @@ def smart_batch_gen(self, batch):
                     )
                 )
             )
-        # continually grab n items from al_psm until list is exhausted
+        # continually grab n items from all_psm until list is exhausted
         while len(all_psm) > 0:
             batch = all_psm[: self.num_pairs]
             all_psm = all_psm[self.num_pairs :]
@@ -1007,20 +1013,21 @@ def smart_batch_gen(self, batch):
             pep_str = list(batch[2])
             indexes = [a[1] for a in batch[3]]
             t_or_d = batch[4]
-            yield (indexes, t_or_d, (encoded_ms, prec_data, pep_str))
+            yield (indexes, t_or_d, pep_str, prec_data, encoded_ms)
 
     def on_predict_epoch_end(self, results) -> None:
         if self.out_writer is None:
             return
-        results = np.array(results, dtype=object).squeeze()
+        results = np.array(results, dtype=object).squeeze((0, 1))
         with open(self.out_writer.filename, "a") as out_f:
             csv_writer = csv.writer(out_f)
-            for index, t_or_d, peptide, score, per_aa_scores in list(
-                zip(*results)
-            ):
-                csv_writer.writerow(
-                    (index, peptide, bool(t_or_d), score, per_aa_scores)
-                )
+            for batch in results:
+                for index, t_or_d, peptide, score, per_aa_scores in list(
+                    zip(*batch)
+                ):
+                    csv_writer.writerow(
+                        (index, peptide, bool(t_or_d), score, per_aa_scores)
+                    )
         out_f.close()
 
 
@@ -1063,10 +1070,9 @@ def calc_match_score(
             assert round(sum(preds).item()) == 1
         aa_scores = []
         for scores, true_index in zip(all_aa_pred, truth_indicies):
-            aa_scores.append(scores[true_index].item())
-        normalized_score = sum(aa_scores) / len(
-            aa_scores
-        )  #! Convert to product and normalization by length, or alternatively use LogSoftmax
+            if true_index != 0:
+                aa_scores.append(scores[true_index].item())
+        normalized_score = sum(aa_scores) / len(aa_scores)
         all_scores.append(normalized_score)
         per_aa_scores.append(aa_scores)
     return all_scores, per_aa_scores