made -unspecified- the default selection for config and split, added metric selection to url, speedup for case where no dataset is selected

app.py

@@ -141,6 +141,7 @@ print(default_only_verified)
 default_dataset = first_query_params.get("dataset", [None])[0]
 default_split = first_query_params.get("split", [None])[0]
 default_config = first_query_params.get("config", [None])[0]
+default_metric = first_query_params.get("metric", [None])[0]
 
 only_verified_results = st.sidebar.checkbox(
     "Filter for Verified Results",
@@ -177,7 +178,9 @@ dataset = st.sidebar.selectbox(
 
 dataframe = dataframe[dataframe.only_verified == only_verified_results]
 
-st.experimental_set_query_params(**{"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task]})
+current_query_params = {"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task]}
+
+st.experimental_set_query_params(**current_query_params)
 
 if dataset != "-any-":
     dataset_df = dataframe[dataframe.dataset == dataset]
@@ -187,8 +190,12 @@ else:
 dataset_df = dataset_df.dropna(axis="columns", how="all")
 
 if len(dataset_df) > 0:
-
-    selectable_configs = list(set(dataset_df["config"]))
+    config_set = set(dataset_df["config"])
+    if "-unspecified-" in config_set:
+        config_set.discard("-unspecified-")
+        selectable_configs = ["-unspecified-"] + list(config_set)
+    else:
+        selectable_configs = list(config_set)
 
     if dataset != "-any-":
         config = st.sidebar.selectbox(
@@ -199,7 +206,12 @@ if len(dataset_df) > 0:
         )
         dataset_df = dataset_df[dataset_df.config == config]
 
-        selectable_splits = list(set(dataset_df["split"]))
+        split_set = set(dataset_df["split"])
+        if "-unspecified-" in split_set:
+            split_set.discard("-unspecified-")
+            selectable_splits = ["-unspecified-"] + list(split_set)
+        else:
+            selectable_splits = list(split_set)
         split = st.sidebar.selectbox(
             "Split",
             selectable_splits,
@@ -207,7 +219,9 @@ if len(dataset_df) > 0:
             help="Filter the results on the current leaderboard by the dataset split. Self-reported results might not report the split, which is why \"-unspecified-\" is an option."
         )
 
-        st.experimental_set_query_params(**{"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task], "config": [config], "split": [split]})
+        current_query_params.update({"config": [config], "split": [split]})
+
+        st.experimental_set_query_params(**current_query_params)
 
         dataset_df = dataset_df[dataset_df.split == split]
 
@@ -220,9 +234,14 @@ if len(dataset_df) > 0:
     sorting_metric = st.sidebar.radio(
         "Sorting Metric",
         selectable_metrics,
+        index=selectable_metrics.index(default_metric) if default_metric in selectable_metrics else 0,
         help="Select the metric to sort the leaderboard by. Click on the metric name in the leaderboard to reverse the sorting order."
     )
 
+    current_query_params.update({"metric": [sorting_metric]})
+
+    st.experimental_set_query_params(**current_query_params)
+
     st.markdown(
         "Please click on the model's name to be redirected to its model card."
     )
@@ -237,7 +256,7 @@ if len(dataset_df) > 0:
 
     if dataset == "-any-":
         st.info(
-            "Note: you haven't chosen a dataset, so the leaderboard is showing the best scoring model for each dataset."
+            "Note: you haven't chosen a dataset, so the leaderboard is showing the best scoring model for a random sample of the datasets available."
         )
 
     # Make the default metric appear right after model names and dataset names
@@ -251,8 +270,8 @@ if len(dataset_df) > 0:
     dataset_df = dataset_df.sort_values(by=cols[sorting_metric_index:], ascending=[metric in ascending_metrics for metric in cols[sorting_metric_index:]])
     dataset_df = dataset_df.replace(np.nan, '-')
 
-    # If dataset is "-any-", only show the best model for each dataset. Otherwise
-    # The leaderboard is way too long and doesn't give the users a feel for all of
+    # If dataset is "-any-", only show the best model for a random sample of 100 datasets.
+    # Otherwise The leaderboard is way too long and doesn't give the users a feel for all of
     # the datasets available for a task.
     if dataset == "-any-":
         filtered_dataset_df_dict = {column: [] for column in dataset_df.columns}
@@ -263,6 +282,7 @@ if len(dataset_df) > 0:
                     filtered_dataset_df_dict[column].append(row[column])
                 seen_datasets.add(row["dataset"])
         dataset_df = pd.DataFrame(filtered_dataset_df_dict)
+        dataset_df = dataset_df.sample(min(100, len(dataset_df)))
 
     # Make the leaderboard
     gb = GridOptionsBuilder.from_dataframe(dataset_df)