add fleurs

app.py

@@ -1,7 +1,7 @@
 import pandas as pd
 import streamlit as st
 from huggingface_hub import HfApi
-from utils import ascending_metrics, metric_ranges, LANGUAGES
+from utils import ascending_metrics, metric_ranges, CV11_LANGUAGES, FLEURS_LANGUAGES
 import numpy as np
 from st_aggrid import AgGrid, GridOptionsBuilder, JsCode
 from os.path import exists
@@ -130,66 +130,69 @@ dataframe = get_data_wrapper()
 
 st.markdown("# 🤗 Whisper Event: Final Leaderboard")
 
+# query params are used to refine the browser URL as more options are selected
 query_params = st.experimental_get_query_params()
 if "first_query_params" not in st.session_state:
     st.session_state.first_query_params = query_params
 first_query_params = st.session_state.first_query_params
 
-default_config = first_query_params.get("config", [None])[0]
-default_metric = first_query_params.get("metric", [None])[0]
-
+# define the scope of the leaderboard
 only_verified_results = False
 task = "automatic-speech-recognition"
-dataset = "mozilla-foundation/common_voice_11_0"
+selectable_datasets = ["mozilla-foundation/common_voice_11_0", "google/fleurs"]
+dataset_mapping = {"mozilla-foundation/common_voice_11_0": "Common Voice 11", "google/fleurs": "FLEURS"}  # get a 'pretty' name for our datasets
 split = "test"
+selectable_metrics = ["wer", "cer"]
+default_metric = selectable_metrics[0]
 
-dataframe = dataframe[dataframe.only_verified == only_verified_results]
-
-current_query_params = {"dataset": [dataset], "only_verified": [int(only_verified_results)], "task": [task],
-                        "split": [split]}
-
-st.experimental_set_query_params(**current_query_params)
-
-dataset_df = dataframe[dataframe.dataset == dataset]
-dataset_df = dataset_df[dataset_df.split == split]
-
-dataset_df = dataset_df.dropna(axis="columns", how="all")
-
+# select dataset from list provided
 dataset = st.sidebar.selectbox(
     "Dataset",
-    [dataset],
-    index=0,
+    selectable_datasets,
+    help="Select a dataset to see the leaderboard!"
 )
+dataset_name = dataset_mapping[dataset]
 
+# slice dataframe to entries of interest
+dataframe = dataframe[dataframe.only_verified == only_verified_results]
+dataset_df = dataframe[dataframe.dataset == dataset]
+dataset_df = dataset_df[dataset_df.split == split]  # hardcoded to "test"
+dataset_df = dataset_df.dropna(axis="columns", how="all")
+
+# get potential dataset configs (languages)
 selectable_configs = list(set(dataset_df["config"]))
 selectable_configs.sort(key=lambda name: name.lower())
-selectable_configs.remove("-unspecified-")
-selectable_configs = [config for config in selectable_configs if config in LANGUAGES]
 
-visual_configs = [f"{config}: {LANGUAGES[config]}" for config in selectable_configs]
+if "-unspecified-" in selectable_configs:
+    selectable_configs.remove("-unspecified-")
+
+if dataset == "mozilla-foundation/common_voice_11_0":
+    selectable_configs = [config for config in selectable_configs if config in CV11_LANGUAGES]
+    visual_configs = [f"{config}: {CV11_LANGUAGES[config]}" for config in selectable_configs]
+elif dataset == "google/fleurs":
+    selectable_configs = [config for config in selectable_configs if config in FLEURS_LANGUAGES]
+    visual_configs = [f"{config}: {FLEURS_LANGUAGES[config]}" for config in selectable_configs]
 
 config = st.sidebar.selectbox(
     "Language",
     visual_configs,
-    index=0,
     help="Filter the results on the current leaderboard by language."
 )
+config, language = config.split(":")
 
-config = config.split(":")[0]
-
-dataset_df = dataset_df[dataset_df.config == config]
-
+# just for show -> we've fixed the split to "test"
 split = st.sidebar.selectbox(
     "Split",
     [split],
     index=0,
+    help="View the results for the `test` split for evaluation performance.",
 )
 
-not_selectable_metrics = ["model_id", "dataset", "split", "config", "pipeline_tag", "only_verified"]
-# also ignore irrelevant ASR metrics
-not_selectable_metrics.extend(["wer_without_norm", "mer"])
+# update browser URL with selections
+current_query_params = {"dataset": [dataset], "config": [config], "split": split}
+st.experimental_set_query_params(**current_query_params)
 
-selectable_metrics = list(filter(lambda column: column not in not_selectable_metrics, dataset_df.columns))
+dataset_df = dataset_df[dataset_df.config == config]
 
 dataset_df = dataset_df.filter(["model_id"] + (["dataset"] if dataset == "-any-" else []) + 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).
@@ -201,12 +204,8 @@ sorting_metric = st.sidebar.radio(
     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(
-    f"This is the leaderboard for {LANGUAGES[config]} ({config})."
+    f"This is the leaderboard for {dataset_name} {language} ({config})."
 )
 
 st.markdown(
@@ -214,7 +213,7 @@ st.markdown(
 )
 
 st.markdown(
-    "Want to beat the leaderboard? Don't see your model here? Simply ..."
+    "Want to beat the leaderboard? Don't see your model here? Ensure..."
 )
 
 # Make the default metric appear right after model names and dataset names
@@ -237,7 +236,7 @@ gb.configure_column(
 )
 
 for name in selectable_metrics:
-    gb.configure_column(name, type=["numericColumn","numberColumnFilter","customNumericFormat"], precision=4, aggFunc='sum')
+    gb.configure_column(name, type=["numericColumn", "numberColumnFilter", "customNumericFormat"], precision=4, aggFunc='sum')
 
 gb.configure_column(
     sorting_metric,