WER & CER Test outputs corrected. (#114)

- unbabel-comet is updated to more recent version.

requirements-dev.txt

@@ -1,7 +1,6 @@
 # Jury metrics
 # Falling back to fork for Windows related issue: https://github.com/google-research/bleurt/issues/49
 bleurt @ git+https://github.com/devrimcavusoglu/bleurt.git
-unbabel-comet @ git+https://github.com/Unbabel/COMET.git@c772b679e20725e6cc79b2107d50594f9ea7a4ae
 
 # for datasets metric test
 math_equivalence @ git+https://github.com/hendrycks/math.git

setup.py

@@ -52,6 +52,7 @@ def add_pywin(reqs: List[str]) -> None:
     "jiwer>=2.3.0",
     "seqeval==1.2.2",
     "sentencepiece==0.1.96",
+    "unbabel-comet>=1.1.2",
 ]
 
 _METRIC_REQUIREMENTS.extend(_PRISM_REQUIREMENTS)

tests/test_data/expected_outputs/metrics/test_cer.json

@@ -31,10 +31,10 @@
     "cer": {
       "score": 0.7272727272727273,
       "overall": {
-        "substitutions": 6.25,
-        "deletions": 5.5,
-        "insertions": 1.5,
-        "hits": 16.25
+        "substitutions": 4.75,
+        "deletions": 6.25,
+        "insertions": 2.25,
+        "hits": 17.0
       }
     }
   },
@@ -44,10 +44,10 @@
     "cer": {
       "score": 0.8484848484848485,
       "overall": {
-        "substitutions": 6.625,
-        "deletions": 3.875,
-        "insertions": 3.625,
-        "hits": 17.5
+        "substitutions": 5.125,
+        "deletions": 4.625,
+        "insertions": 4.375,
+        "hits": 18.25
       }
     }
   }

tests/test_data/expected_outputs/metrics/test_wer.json

@@ -31,10 +31,10 @@
     "wer": {
       "score": 1.0,
       "overall": {
-        "substitutions": 3.25,
-        "deletions": 0.5,
-        "insertions": 0.25,
-        "hits": 2.0
+        "substitutions": 2.75,
+        "deletions": 0.75,
+        "insertions": 0.5,
+        "hits": 2.25
       }
     }
   },
@@ -44,10 +44,10 @@
     "wer": {
       "score": 1.2,
       "overall": {
-        "substitutions": 3.0,
-        "deletions": 0.375,
-        "insertions": 0.625,
-        "hits": 2.375
+        "substitutions": 2.75,
+        "deletions": 0.5,
+        "insertions": 0.75,
+        "hits": 2.5
       }
     }
   }