fix surprisal indexing issues

syntaxgym.py

@@ -203,6 +203,7 @@ class SyntaxGymMetric(datasets.Metric):
 
         # TODO: check this logic
         tt = expected_ids.unsqueeze(2)
+        # reindexed surprisals: B * (T - 1)
         surprisals = torch.gather(surps_shifted, 2, expected_ids.unsqueeze(2)) \
             .squeeze(2)
         # This is the original, which works but not with multiple axes in expected_ids
@@ -220,8 +221,11 @@ class SyntaxGymMetric(datasets.Metric):
         for i, (i_cond, i_inputs) in enumerate(zip(condition_names, input_ids)):
             for region_number, region_tokens in region2tokens[i_cond].items():
                 for token in region_tokens:
-                    if token < surprisals.shape[1]:
-                        region_totals[i_cond][region_number] += surprisals[i, token]
+                    if token == 0:
+                        # surprisal not defined. pass.
+                        continue
+                    elif token <= surprisals.shape[1]:
+                        region_totals[i_cond][region_number] += surprisals[i, token - 1]
                     else:
                         # TODO don't think this is an issue, just should clean
                         # up the aggregation output