improve missing repr

app.py

@@ -38,17 +38,18 @@ class Result:
     model_type: Literal["pretrained", "fine-tuned", "instruction-tuned", "RL-tuned"]
     dutch_coverage: Literal["none", "pretrained", "fine-tuned"]
     num_parameters: int
-    arc: float = field(default=0.0)
-    average: float = field(default=0.0, init=False)
-    hellaswag: float = field(default=0.0)
-    mmlu: float = field(default=0.0)
-    truthfulqa: float = field(default=0.0)
+    arc: float = field(default=np.nan)
+    average: float = field(default=np.nan, init=False)
+    hellaswag: float = field(default=np.nan)
+    mmlu: float = field(default=np.nan)
+    truthfulqa: float = field(default=np.nan)
     num_parameters_kmb: str = field(init=False)
 
     def __post_init__(self):
         if self.model_type not in ["pretrained", "fine-tuned", "instruction-tuned", "RL-tuned", "not-given"]:
             raise ValueError(
-                f"Model type {self.model_type} must be one of 'pretrained', 'fine-tuned', 'instruction-tuned', 'RL-tuned', 'not-given"
+                f"Model type {self.model_type} must be one of 'pretrained', 'fine-tuned',"
+                f" 'instruction-tuned', 'RL-tuned', 'not-given"
             )
         if self.dutch_coverage not in ["none", "pretrained", "fine-tuned", "not-given"]:
             raise ValueError(
@@ -60,7 +61,10 @@ class Result:
             if task_name not in field_names:
                 raise ValueError(f"Task name {task_name} not found in Result class fields so cannot create DataFrame")
 
-        self.average = (self.arc + self.hellaswag + self.mmlu + self.truthfulqa) / 4
+        if any([np.isnan(getattr(self, task_name)) for task_name in TASK_METRICS]):
+            self.average = np.nan
+        else:
+            self.average = sum([getattr(self, task_name) for task_name in TASK_METRICS]) / 4
         self.num_parameters_kmb = convert_number_to_kmb(self.num_parameters)
 
 
@@ -145,23 +149,22 @@ class ResultSet:
         df = pd.DataFrame(data)
         df = df.sort_values(by=self.column_names["average"], ascending=False)
         number_cols = [col for attr, col in self.column_names.items() if attr in TASK_METRICS or attr == "average"]
-        styler = df.style.format("{:.2f}", subset=number_cols)
+        styler = df.style.format("{:.2f}", subset=number_cols, na_rep="<missing>")
 
         def highlight_max(col):
             return np.where(col == np.nanmax(col.to_numpy()), "font-weight: bold;", None)
 
         styler = styler.apply(highlight_max, axis=0, subset=number_cols)
-
         num_params_col = self.column_names["num_parameters"]
         styler = styler.format(convert_number_to_kmb, subset=num_params_col)
-
+        styler.set_caption("Leaderboard on Dutch benchmarks.")
         styler = styler.hide()
         return styler
 
     @cached_property
     def latex_df(self) -> Styler:
         number_cols = [col for attr, col in self.column_names.items() if attr in TASK_METRICS or attr == "average"]
-        styler = self.df.style.format("{:.2f}", subset=number_cols)
+        styler = self.df.style.format("{:.2f}", subset=number_cols, na_rep="<missing>")
 
         def highlight_max(col):
             return np.where(col == np.nanmax(col.to_numpy()), "font-weight: bold;", None)
@@ -169,6 +172,7 @@ class ResultSet:
         styler = styler.apply(highlight_max, axis=0, subset=number_cols)
         num_params_col = self.column_names["num_parameters"]
         styler = styler.format(convert_number_to_kmb, subset=num_params_col)
+        styler.set_caption("Leaderboard on Dutch benchmarks.")
         styler = styler.hide()
         return styler
 
@@ -244,7 +248,8 @@ with gr.Blocks() as demo:
 
     gr.Markdown(
         f"## Leaderboard\nOnly representative for the Dutch version (`*_nl`) of the benchmarks!"
-        " All models have been benchmarked in 8-bit."
+        " All models have been benchmarked in 8-bit. `<missing>` values indicate that those benchmarks are still"
+        " pending."
     )
 
     results = collect_results()