Plot with plotly

requirements.txt

@@ -0,0 +1,2 @@
+plotly
+

src/results.py

@@ -157,8 +157,11 @@ def plot_results(task, *dfs):
                 and (col.endswith("acc,none") or col.endswith("acc_norm,none") or col.endswith("exact_match,none"))
             ]
         ]
+        tasks = {key: tupl[0] for key, tupl in TASKS.items()}
+        tasks["leaderboard_math"] = tasks["leaderboard_math_hard"]
+        subtasks = {tupl[1]: tupl[0] for tupl in constants.SUBTASKS.get(task, [])}
         if task == "All":
-            df = df[[col for col in df.columns if col.split(".")[1] in TASKS]]
+            df = df[[col for col in df.columns if col.split(".")[1] in tasks]]
             # - IFEval: Calculate average of both strict accuracies
             ifeval_mean = df[
                 [
@@ -170,19 +173,19 @@ def plot_results(task, *dfs):
             loc = df.columns.get_loc("results.leaderboard_math_hard.exact_match,none")
             df.insert(loc - 1, "results.leaderboard_ifeval", ifeval_mean)
             # Rename
-            df = df.rename(columns=lambda col: TASKS[col.split(".")[1]][0])
+            df = df.rename(columns=lambda col: tasks[col.split(".")[1]])
         else:
             df = df[[col for col in df.columns if col.startswith(f"results.{task}")]]
-            tasks = {key: tupl[0] for key, tupl in TASKS.items()}
-            subtasks = {tupl[1]: tupl[0] for value in constants.SUBTASKS.values() for tupl in value}
-            subtasks = {**tasks, **subtasks}
             # - IFEval: Return 4 accuracies
             if task == "leaderboard_ifeval":
                 df = df.rename(columns=lambda col: col.split(".")[2].removesuffix(",none"))
             else:
-                df = df.rename(columns=lambda col: subtasks[col.split(".")[1]])
-        ax = df.T.rename_axis(columns="Models").plot(kind="bar", ylabel="Scores", rot=45, figsize=(18, 6))
-        fig = ax.get_figure()
-        fig.autofmt_xdate(rotation=45)
-        fig.tight_layout()
+                df = df.rename(columns=lambda col: tasks.get(col.split(".")[1], subtasks.get(col.split(".")[1])))
+        fig = df.T.rename_axis(columns="Model").plot.bar(
+            backend="plotly",
+            barmode="group",
+            labels={"index": "Benchmark" if task == "All" else "Subtask", "value": "Score"},
+            color_discrete_sequence=["#FF9D00", "#32343D"],
+        )
+        fig.update_yaxes(range=[0, 1])
         return fig