fixed cuttoff issue for wide leaderboards, made leaderboard data updating asynchronous, made streamlit set the url to match the selected dataset

app.py

@@ -1,4 +1,3 @@
-import requests
 import pandas as pd
 from tqdm.auto import tqdm
 import streamlit as st
@@ -6,17 +5,9 @@ from huggingface_hub import HfApi, hf_hub_download
 from huggingface_hub.repocard import metadata_load
 from ascending_metrics import ascending_metrics
 import numpy as np
-
-
-def make_clickable(model_name):
-    link = "https://huggingface.co/" + model_name
-    return f'<a target="_blank" href="{link}">{model_name}</a>'
-
-def make_bold(value):
-    return f'<b>{value}</b>'
-
-def make_string(value):
-    return str(value)
+from st_aggrid import AgGrid, GridOptionsBuilder, JsCode
+from os.path import exists
+import threading
 
 
 def get_model_ids():
@@ -71,24 +62,42 @@ def parse_metrics_rows(meta):
                 row[type] = value
         yield row
 
+@st.cache(ttl=3600)
+def get_data_wrapper():
 
-@st.cache(ttl=86400)
-def get_data():
-    data = []
-    model_ids = get_model_ids()
-    for model_id in tqdm(model_ids):
-        meta = get_metadata(model_id)
-        if meta is None:
-            continue
-        for row in parse_metrics_rows(meta):
-            if row is None:
+    def get_data():
+        data = []
+        model_ids = get_model_ids()
+        for model_id in tqdm(model_ids):
+            meta = get_metadata(model_id)
+            if meta is None:
                 continue
-            row["model_id"] = model_id
-            data.append(row)
-    return pd.DataFrame.from_records(data)
+            for row in parse_metrics_rows(meta):
+                if row is None:
+                    continue
+                row["model_id"] = model_id
+                data.append(row)
+        dataframe = pd.DataFrame.from_records(data)
+        dataframe.to_pickle("cache.pkl")
+
+    if exists("cache.pkl"):
+        # If we have saved the results previously, call an asynchronous process
+        # to fetch the results and update the saved file. Don't make users wait
+        # while we fetch the new results. Instead, display the old results for
+        # now. The new results should be loaded when this method
+        # is called again.
+        dataframe = pd.read_pickle("cache.pkl")
+        t = threading.Thread(name='get_data procs', target=get_data)
+        t.start()
+    else:
+        # We have to make the users wait during the first startup of this app.
+        get_data()
+        dataframe = pd.read_pickle("cache.pkl")
+
+    return dataframe
+
+dataframe = get_data_wrapper()
 
-
-dataframe = get_data()
 selectable_datasets = list(set(dataframe.dataset.tolist()))
 
 st.markdown("# 🤗 Leaderboards")
@@ -104,19 +113,20 @@ dataset = st.sidebar.selectbox(
     selectable_datasets,
     index=selectable_datasets.index(default_dataset),
 )
+st.experimental_set_query_params(**{"dataset": [dataset]})
 
 dataset_df = dataframe[dataframe.dataset == dataset]
 dataset_df = dataset_df.dropna(axis="columns", how="all")
 
 selectable_metrics = list(filter(lambda column: column not in ("model_id", "dataset"), dataset_df.columns))
-metric = st.sidebar.radio(
-    "Sorting Metric",
+default_metric = st.sidebar.radio(
+    "Default Metric",
     selectable_metrics,
 )
 
 dataset_df = dataset_df.filter(["model_id"] + selectable_metrics)
 dataset_df = dataset_df.dropna(thresh=2)  # Want at least two non-na values (one for model_id and one for a metric).
-dataset_df = dataset_df.sort_values(by=metric, ascending=metric in ascending_metrics)
+dataset_df = dataset_df.sort_values(by=default_metric, ascending=default_metric in ascending_metrics)
 dataset_df = dataset_df.replace(np.nan, '-')
 
 st.markdown(
@@ -127,30 +137,25 @@ st.markdown(
     "Want to beat the leaderboard? Don't see your model here? Simply request an automatic evaluation [here](https://huggingface.co/spaces/autoevaluate/autoevaluate)."
 )
 
-# display the model ranks
-dataset_df = dataset_df.reset_index(drop=True)
-dataset_df.index += 1
-
-# turn the model ids into clickable links
-dataset_df["model_id"] = dataset_df["model_id"].apply(make_clickable)
-dataset_df[metric] = dataset_df[metric].apply(make_bold)
-for other_metric in selectable_metrics:
-    dataset_df[other_metric] = dataset_df[other_metric].apply(make_string)
-
-# Make the selected metric appear right after model names
+# Make the default metric appear right after model names
 cols = dataset_df.columns.tolist()
-cols.remove(metric)
-cols = cols[:1] + [metric] + cols[1:]
+cols.remove(default_metric)
+cols = cols[:1] + [default_metric] + cols[1:]
 dataset_df = dataset_df[cols]
 
-# Highlight selected metric
-def highlight_cols(s):
-    huggingface_yellow = "#FFD21E"
-    return "background-color: %s" % huggingface_yellow
+# Make the leaderboard
+gb = GridOptionsBuilder.from_dataframe(dataset_df)
+gb.configure_column(
+    "model_id",
+    cellRenderer=JsCode('''function(params) {return '<a target="_blank" href="https://huggingface.co/'+params.value+'">'+params.value+'</a>'}'''),
+)
+for name in selectable_metrics:
+    gb.configure_column(name, type=["numericColumn","numberColumnFilter","customNumericFormat"], precision=2, aggFunc='sum')
 
-dataset_df = dataset_df.style.applymap(highlight_cols, subset=pd.IndexSlice[metric])
+gb.configure_column(
+    default_metric,
+    cellStyle=JsCode('''function(params) { return {'backgroundColor': '#FFD21E'}}''')
+)
 
-# Turn table into html
-table_html = dataset_df.to_html(escape=False)
-table_html = table_html.replace("<th>", '<th align="left">')  # left-align the headers
-st.write(table_html, unsafe_allow_html=True)
+go = gb.build()
+AgGrid(dataset_df, gridOptions=go, allow_unsafe_jscode=True, fit_columns_on_grid_load=True)

requirements.txt

@@ -2,4 +2,5 @@ pandas
 tqdm
 streamlit
 huggingface_hub
-numpy
+numpy
+streamlit-aggrid