Improved InstanceScores summary to be readible and in decent width (#1847)

Author: yoavkatz

examples/key_value_extraction_evaluation.py

@@ -71,7 +71,7 @@ def text_to_image(text: str):
 print(json.dumps(results.instance_scores[0]["source"], indent=4))
 
 print("Instance Results:")
-print(results.instance_scores)
+print(results.instance_scores.summary)
 
 print("Global Results:")
 print(results.global_scores.summary)

src/unitxt/metric_utils.py

@@ -687,7 +687,11 @@ def to_df(self, flatten=True, columns=None):
     def _to_markdown(self, df, max_col_width=30, **kwargs):
         def wrap_column(series, max_width=30):
             """Wraps string values in a Pandas Series to a maximum width."""
-            return series.apply(lambda x: textwrap.fill(str(x), width=max_width))
+            return series.apply(
+                lambda x: "\n".join(
+                    textwrap.fill(line, width=max_width) for line in str(x).splitlines()
+                )
+            )
 
         wrapped_df = df.copy()
         for col in wrapped_df.columns:
@@ -699,22 +703,24 @@ def to_markdown(self, flatten=True, columns=None, max_col_width=30, **kwargs):
 
     @property
     def summary(self):
-        return self._to_markdown(
-            self.to_df()
-            .head()
-            .drop(
-                columns=[
-                    "metadata",
-                    "media",
-                    "groups",
-                    "subset",
-                    "demos",
-                    "metrics",
-                    "postprocessors",
-                ],
-                errors="ignore",
-            )
+        df = self.to_df(
+            flatten=False,
+            columns=[
+                "source",
+                "prediction",
+                "processed_prediction",
+                "references",
+                "processed_references",
+                "score",
+            ],
+        ).head()
+        df["score_name"] = df["score"].apply(lambda x: x["instance"]["score_name"])
+        df["all_scores"] = df["score"].apply(
+            lambda x: "\n".join(f"{k}: {v}" for k, v in x["instance"].items())
         )
+        df["score"] = df["score"].apply(lambda x: x["instance"]["score"])
+
+        return self._to_markdown(df)
 
     def __repr__(self):
         return to_pretty_string(self, float_format=".2g")