add visualization

app.py

@@ -1,5 +1,4 @@
 import json
-from collections import defaultdict
 from dataclasses import dataclass, field, fields
 from functools import cached_property
 from pathlib import Path
@@ -10,6 +9,7 @@ import pandas as pd
 import gradio as gr
 from pandas import DataFrame
 from pandas.io.formats.style import Styler
+import plotly.graph_objects as go
 
 from content import *
 
@@ -176,6 +176,51 @@ class ResultSet:
         styler = styler.hide()
         return styler
 
+    @cached_property
+    def viz_checkboxes(self):
+        model_col_name = self.column_names["short_name"]
+        avg_col = self.column_names["average"]
+        top3_models = self.df.sort_values(by=avg_col, ascending=False)[model_col_name].tolist()[:3]
+        return gr.CheckboxGroup(self.df[model_col_name].tolist(), label="Models", value=top3_models)
+
+    def plot(self, model_names: list[str]):
+        if not model_names:
+            return None
+
+        # Only get task columns and model name
+        task_columns = [col for attr, col in self.column_names.items() if attr in TASK_METRICS or attr == "short_name"]
+        df = self.df[task_columns]
+
+        # Rename the columns to the task names
+        reversed_col_names = {v: k for k, v in self.column_names.items() if v != "Model"}
+        df = df.rename(columns=reversed_col_names)
+
+        # Only keep the selected models
+        df = df[df["Model"].isin(model_names)]
+
+        # Melt the dataframe to long format
+        df = df.melt(id_vars=["Model"], var_name="Task", value_name="Score").sort_values(by="Task")
+
+        # Populate figure
+        fig = go.Figure()
+        for model_name in model_names:
+            model_df = df[df["Model"] == model_name]
+            scores = model_df["Score"].tolist()
+            tasks = model_df["Task"].tolist()
+
+            # Repeat the first point at the end to close the lines
+            # Cf. https://community.plotly.com/t/closing-line-for-radar-cart-and-popup-window-on-chart-radar/47711/4
+            scores.append(scores[0])
+            tasks.append(tasks[0])
+
+            fig.add_trace(go.Scatterpolar(r=scores, theta=tasks, name=model_name))
+
+        fig.update_layout(
+            title="Model performance on Dutch benchmarks",
+        )
+
+        return fig
+
 
 def convert_number_to_kmb(number: int) -> str:
     """
@@ -282,6 +327,16 @@ with gr.Blocks() as demo:
     gr.Markdown("## LaTeX")
     gr.Code(results.latex_df.to_latex(convert_css=True))
 
+    gr.Markdown("## Visualization")
+    with gr.Row():
+        with gr.Column():
+            buttons = results.viz_checkboxes
+
+        with gr.Column(scale=2):
+            plot = gr.Plot(container=True)
+            buttons.change(results.plot, inputs=buttons, outputs=[plot])
+            demo.load(results.plot, inputs=buttons, outputs=[plot])
+
     gr.Markdown(DISCLAIMER, elem_classes="markdown-text")
     gr.Markdown(CREDIT, elem_classes="markdown-text")
     gr.Markdown(CITATION, elem_classes="markdown-text")