Update app.py

app.py

@@ -1,7 +1,6 @@
 import pandas as pd
 import streamlit as st
-from huggingface_hub import HfApi, hf_hub_download
-from huggingface_hub.repocard import metadata_load
+from huggingface_hub import HfApi
 from utils import ascending_metrics, metric_ranges
 import numpy as np
 from st_aggrid import AgGrid, GridOptionsBuilder, JsCode
@@ -80,9 +79,9 @@ def parse_metrics_rows(meta, only_verified=False):
             continue
         yield row
 
+
 @st.cache(ttl=0)
 def get_data_wrapper():
-
     def get_data(dataframe=None, verified_dataframe=None):
         data = []
         verified_data = []
@@ -126,107 +125,69 @@ def get_data_wrapper():
 
     return dataframe
 
+
 dataframe = get_data_wrapper()
 
-st.markdown("# 🤗 Leaderboards")
+st.markdown("# 🤗 Whisper Event: Final Leaderboard")
 
 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_task = first_query_params.get("task", [None])[0]
-default_only_verified = bool(int(first_query_params.get("only_verified", [0])[0]))
-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",
-    value=default_only_verified,
-    help="Select this checkbox if you want to see only results produced by the Hugging Face model evaluator, and no self-reported results."
-)
-
-selectable_tasks = list(set(dataframe.pipeline_tag))
-if None in selectable_tasks:
-    selectable_tasks.remove(None)
-selectable_tasks.sort(key=lambda name: name.lower())
-selectable_tasks = ["-any-"] + selectable_tasks
-
-task = st.sidebar.selectbox(
-    "Task",
-    selectable_tasks,
-    index=(selectable_tasks).index(default_task) if default_task in selectable_tasks else 0,
-    help="Filter the selectable datasets by task. Leave as \"-any-\" to see all selectable datasets."
-)
-
-if task != "-any-":
-    dataframe = dataframe[dataframe.pipeline_tag == task]
-
-selectable_datasets = ["-any-"] + sorted(list(set(dataframe.dataset.tolist())), key=lambda name: name.lower())
-if "" in selectable_datasets:
-    selectable_datasets.remove("")
-
-dataset = st.sidebar.selectbox(
-    "Dataset",
-    selectable_datasets,
-    index=selectable_datasets.index(default_dataset) if default_dataset in selectable_datasets else 0,
-    help="Select a dataset to see the leaderboard!"
-)
+only_verified_results = False
+task = "automatic-speech-recognition"
+dataset = "mozilla-foundation/common_voice_11_0"
+split = "test"
 
 dataframe = dataframe[dataframe.only_verified == only_verified_results]
 
-current_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],
+                        "split": [split]}
 
 st.experimental_set_query_params(**current_query_params)
 
-if dataset != "-any-":
-    dataset_df = dataframe[dataframe.dataset == dataset]
-else:
-    dataset_df = dataframe
+dataset_df = dataframe[dataframe.dataset == dataset]
+dataset_df = dataset_df[dataset_df.split == split]
 
 dataset_df = dataset_df.dropna(axis="columns", how="all")
 
+selectable_datasets = [dataset]
+dataset = st.sidebar.selectbox(
+    "Dataset",
+    selectable_datasets,
+    index=0,
+)
+
 if len(dataset_df) > 0:
     selectable_configs = list(set(dataset_df["config"]))
     selectable_configs.sort(key=lambda name: name.lower())
 
-    if "-unspecified-" in selectable_configs:
-        selectable_configs.remove("-unspecified-")
-        selectable_configs = ["-unspecified-"] + selectable_configs
+    selectable_configs.remove("-unspecified-")
 
     if dataset != "-any-":
         config = st.sidebar.selectbox(
-            "Config",
+            "Language",
             selectable_configs,
-            index=selectable_configs.index(default_config) if default_config in selectable_configs else 0,
-            help="Filter the results on the current leaderboard by the dataset config. Self-reported results might not report the config, which is why \"-unspecified-\" is an option."
+            index=0,
+            help="Filter the results on the current leaderboard by language."
         )
         dataset_df = dataset_df[dataset_df.config == config]
 
-        selectable_splits = list(set(dataset_df["split"]))
-        selectable_splits.sort(key=lambda name: name.lower())
-
-        if "-unspecified-" in selectable_splits:
-            selectable_splits.remove("-unspecified-")
-            selectable_splits = ["-unspecified-"] + selectable_splits
-
+        selectable_splits = [split]
         split = st.sidebar.selectbox(
             "Split",
             selectable_splits,
-            index=selectable_splits.index(default_split) if default_split in selectable_splits else 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."
+            index=0,
         )
 
-        current_query_params.update({"config": [config], "split": [split]})
-
-        st.experimental_set_query_params(**current_query_params)
-
-        dataset_df = dataset_df[dataset_df.split == split]
-
     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"])
+
     selectable_metrics = list(filter(lambda column: column not in not_selectable_metrics, dataset_df.columns))
 
     dataset_df = dataset_df.filter(["model_id"] + (["dataset"] if dataset == "-any-" else []) + selectable_metrics)
@@ -248,18 +209,9 @@ if len(dataset_df) > 0:
     )
 
     st.markdown(
-        "Want to beat the leaderboard? Don't see your model here? Simply request an automatic evaluation [here](https://huggingface.co/spaces/autoevaluate/model-evaluator)."
-    )
-
-    st.markdown(
-        "If you do not see your self-reported results here, ensure that your results are in the expected range for all metrics. E.g., accuracy is 0-1, not 0-100."
+        "Want to beat the leaderboard? Don't see your model here? Simply ..."
     )
 
-    if dataset == "-any-":
-        st.info(
-            "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
     cols = dataset_df.columns.tolist()
     cols.remove(sorting_metric)
@@ -313,4 +265,4 @@ if len(dataset_df) > 0:
 else:
     st.markdown(
         "No " + ("verified" if only_verified_results else "unverified") + " results to display. Try toggling the verified results filter."
-    )
+    )