Update helper.py

helper.py

@@ -30,21 +30,26 @@ TASKS_LIST={
     'pos':'Part-of-Speech Tagging',
 }
 CLUSTERS = {
-    "Text Classification": [
+    "Text Classification Tasks": [
         'xlni', 'lid', 'news', 'sentiment', 'topic',
     ],
-    "Text Generation": [
+    "Text Generation Tasks": [
         'mt_eng2xx', 'mt_fra2xx', 'mt_xx2xx', 'paraphrase', 'summary', 'title',
     ],
-    "MCCR": [
+    "MCCR Tasks": [
         'mmlu', 'mgsm', 'belebele', 'squad_qa',
     ],
-    "Tokens": [
+    "Tokens Level Tasks": [
         'ner', 'phrase', 'pos',
     ],
 }
 ALL_TASKS = [t for cluster in CLUSTERS.values() for t in cluster]
-
+# This dictionary maps each task ID to its parent cluster name
+TASK_TO_CLUSTER_MAP = {
+    task: cluster_name
+    for cluster_name, tasks in CLUSTERS.items()
+    for task in tasks
+}
 # ===== Authenticate and Load Data From Private HF Repo =====
 
 def load_private_leaderboard_df():
@@ -89,9 +94,9 @@ LANG_ISO2NAME = {
     'som': 'Somali',
     'pcm': 'Nigerian Pidgin',
     'teo': 'Teso',
-    'nyn': 'Nyankore/Nyankole',
+    'nyn': 'Nyankore',# (Nyankole)',
     'lgg': 'Lugbara',
-    'bem': 'Bemba/Chibemba',
+    'bem': 'Bemba',# (Chibemba)',
     'tsn': 'Tswana',
     'bbj': 'Ghomálá',
     'mos': 'Moore',
@@ -101,11 +106,11 @@ LANG_ISO2NAME = {
     'nso': 'Sepedi',
     'tso': 'Tsonga',
     'fuv': 'Fulfude Nigeria',
-    'gaz': 'Oromo, West Central',
+    'gaz': 'Oromo', #, West Central',
     'kea': 'Kabuverdianu',
     'nya': 'Nyanja',
     'ssw': 'Swati',
-    'luo': 'Dholuo/Luo',
+    'luo': 'Dholuo',# (Luo)',
     'ven': 'Venda',
     'kir':"Kirundi",
 }
@@ -120,7 +125,100 @@ def build_langname_to_isos(iso2name):
 LANGNAME2ISOS = build_langname_to_isos(LANG_ISO2NAME)
 #show only African langs
 LANG_NAME_LIST = sorted([lang for lang in LANGNAME2ISOS.keys() if lang not in ['eng', 'fra', 'English', 'French']])
+# TASK_NAME_LIST = sorted(list(TASKS_LIST.values()))
+# Create a list of choices in the format "Task Name (id)"
+TASK_NAME_LIST = sorted([f"{name} ({key})" for key, name in TASKS_LIST.items()])
+TASK_NAME2KEY = {v: k for k, v in TASKS_LIST.items()}
+
+def get_model_table(model_name):
+    """
+    Generates a performance table for a specific model, showing cluster, task, and score.
+    The table is sorted by Cluster and then by Task Name.
+    """
+    # Filter for the selected model and only 'main' leaderboard entries
+    model_df = all_df[(all_df['model'] == model_name) & (all_df['leaderboard'] == 'main')].copy()
+
+    if model_df.empty:
+        return pd.DataFrame([{"Info": f"No 'main' leaderboard data available for the model: {model_name}"}])
+
+    # --- NEW: Add the Cluster Name column using the map ---
+    model_df['Cluster'] = model_df['task'].map(TASK_TO_CLUSTER_MAP)
+    
+    # Create other descriptive columns
+    model_df['Task Name'] = model_df['task'].map(TASKS_LIST)
+    model_df['Metric'] = model_df['metric'].map(metrics_list)
+    model_df['Score'] = model_df['score'].apply(lambda x: f"{x:.2f}" if pd.notna(x) else "---")
+
+    # --- MODIFIED: Select the new 'Cluster' column for the final table ---
+    table = model_df[['Cluster', 'Task Name', 'task', 'Metric', 'Score']].rename(columns={'task': 'Task ID'})
+
+    # --- MODIFIED: Sort by Cluster first, then by Task Name ---
+    table = table.sort_values(by=['Cluster', 'Task Name']).reset_index(drop=True)
+
+    # Handle cases where a task might not be in a cluster
+    table['Cluster'].fillna('Uncategorized', inplace=True)
+
+    return table
+    
+def get_task_leaderboard(task_key):
+    """
+    Generates a leaderboard for a specific task, showing model performance across all languages.
+    """
+    # Filter the main DataFrame for the selected task
+    task_df = all_df[all_df['task'] == task_key].copy()
+
+    if task_df.empty:
+        return pd.DataFrame([{"Info": f"No data available for the task: {TASKS_LIST.get(task_key, task_key)}"}])
+
+    # Get the metric for this task to display later
+    metric_name = metrics_list.get(task_df['metric'].iloc[0], '')
 
+    # Create a user-friendly column name for each language/leaderboard
+    def make_lang_col(row):
+        lb = row['leaderboard']
+        if lb == 'main':
+            # Skip the 'main' leaderboard for task-specific views as it's an aggregate
+            return None
+        if '-' in lb:
+            pair_lang = lb.split('-')
+            # Handles cases where an ISO code might not be in our map
+            src_lang = LANG_ISO2NAME.get(pair_lang[0], pair_lang[0])
+            tgt_lang = LANG_ISO2NAME.get(pair_lang[1], pair_lang[1])
+            return f"{src_lang} to {tgt_lang}"
+        else:
+            return LANG_ISO2NAME.get(lb, lb)
+    if task_key not in ['lid']:
+        task_df['lang_col'] = task_df.apply(make_lang_col, axis=1)
+        task_df.dropna(subset=['lang_col'], inplace=True) # Remove rows where lang_col is None
+    
+        if task_df.empty:
+            return pd.DataFrame([{"Info": f"No language-specific data for the task: {TASKS_LIST.get(task_key, task_key)}"}])
+
+        # Pivot the table to have models as rows and languages as columns
+        table = task_df.pivot_table(index='model', columns='lang_col', values='score', aggfunc='mean').reset_index()
+    else:
+        table = task_df.pivot_table(index='model', columns='task', values='score', aggfunc='mean').reset_index()
+    
+    score_cols = [col for col in table.columns if col != 'model']
+    for col in score_cols:
+        table[col] = table[col].apply(lambda x: f"{x:.2f}" if isinstance(x, (int, float)) else x)
+    main_score_map = all_df[(all_df['task'] == task_key) & (all_df['leaderboard'] == 'main')].set_index('model')['score']
+    table.insert(1, 'Task Score', table['model'].map(main_score_map).apply(lambda x: f"{x:.2f}" if pd.notna(x) else "---"))
+
+    # Add ranking medals based on the "Task Score"
+    table = add_medals_to_models(table, score_col="Task Score")
+    
+    # Rename columns to be more descriptive, including the metric
+    # rename_cols = {col: f"{col}<br>Metric: {metric_name}" for col in score_cols}
+    if task_key in ['belebele', 'ner', 'mgsm', 'mmlu']:
+        # rename_cols = {col: f"<div class='rotate_div'><br>{next(iter(LANGNAME2ISOS.get(col)))}</div>" for col in score_cols}
+        rename_cols = {col: f"<div class='rotate_div'><br>{col}</div>" for col in score_cols}
+    else:
+        rename_cols = {col: f"{col}" for col in score_cols}
+    table.rename(columns=rename_cols, inplace=True)
+
+    return table
+    
 def get_task_metric_map(df):
     mapping = {}
     for _, row in df.iterrows():
@@ -208,60 +306,17 @@ def df_to_html(df, col_minwidth=90, col_maxwidth=140, model_col_width=400):
     # Remove any column whose name contains "task"
     drop_cols = [col for col in df.columns if "task" in col]
     df = df.drop(columns=drop_cols, errors="ignore")
-    df.columns.name = None
-    html=""
-    # html = f"""
-    # <style>
-    #     .gradio-container-5-34-1 .prose table {{
-    #         border-top: 2px solid #dca02a;
-    #         border-bottom: 2px solid #dca02a;
-    #         margin-bottom:20px;
-    #         margin-left: auto;
-    #         margin-right: auto;
-    #         width: 100%;
-    #         border-collapse: collapse;
-    #         table-layout: fixed;
-    #     }}
-    #     .gradio-container-5-34-1 .prose thead tr {{
-    #         background: #fffbe9;
-    #         border-bottom: 2px solid #dca02a;
-    #     }}
-    #     .gradio-container-5-34-1 .prose th {{
-    #         color: #7d3561;
-    #         font-weight: bold;
-    #         font-size: 20px;
-    #         background: #fffbe9;
-    #         padding: 8px 5px;
-    #         vertical-align: middle;
-    #         border: 0px solid #e0e0e0;
-    #     }}
-    #     td {{
-    #         font-size: 18px;
-    #         padding: 8px 5px;
-    #         border: 0px solid #e0e0e0;
-    #         vertical-align: middle;
-    #     }}
-    #     th:first-child, td:first-child {{
-    #         min-width: {model_col_width}px !important;
-    #         max-width: {model_col_width}px !important;
-    #         width: {model_col_width}px !important;
-    #         text-align: left !important;
-    #     }}
-    #     th:not(:first-child), td:not(:first-child) {{
-    #         min-width: {col_minwidth}px;
-    #         max-width: {col_maxwidth}px;
-    #         width: auto;
-    #         text-align: center;
-    #     }}
-    # </style>
-    # """
-    html += df.to_html(index=False, escape=False)
+    df.columns.name = None   
+    html = df.to_html(index=False, escape=False)
     return html
 
 
 
 cluster_tabs, main_overall_tab, all_df, metric_map = load_leaderboards()
 
+# Get the list of unique model names for the new dropdown
+MODEL_NAME_LIST = sorted(all_df['model'].unique()) if not all_df.empty else []
+
 def get_lang_table(lang_name):
     iso_codes = LANGNAME2ISOS.get(lang_name, [])
     if not iso_codes: