GSK-2836-change-ui-based-on-QA-fix-splits-bug (#119)

- fix ui and splits not updated bug (261e3ff7b4e4b2a721516698fc03bde46ee5990c)
- clean code (07461b51a6c9c15efc7988a8ca6d2127d88d4359)

app_debug.py

@@ -74,12 +74,12 @@ def get_demo():
             value=get_queue_status,
             every=5,
         )
-    with gr.Accordion(label="Log Files", open=False):
-        with gr.Row():
-            gr.Files(value=get_log_files, label="Log Files", every=10)
+    with gr.Accordion(label="Log Files", open=True):
         with gr.Row():
             gr.Textbox(
                 value=get_logs_file, every=0.5, lines=10, visible=True, label="Current Log File"
             )
+        with gr.Row():
+            gr.Files(value=get_log_files, label="Log Files", every=10)
     with gr.Accordion(label="Config Files", open=False):
         gr.Files(value=get_config_files, label="Config Files", every=10)

app_leaderboard.py

@@ -88,11 +88,29 @@ def get_demo(leaderboard_tab):
     dataset_ids = get_dataset_ids(records)
 
     column_names = records.columns.tolist()
-    default_columns = ["model_id", "dataset_id", "total_issues", "report_link"]
+    issue_columns = column_names[:11]
+    info_columns = column_names[15:]
+    default_columns = ["dataset_id", "total_issues", "report_link"]
     default_df = records[default_columns]  # extract columns selected
     types = get_types(default_df)
     display_df = get_display_df(default_df)  # the styled dataframe to display
 
+    with gr.Row():
+        with gr.Column():
+          issue_columns_select = gr.CheckboxGroup(
+              label="Issue Columns",
+              choices=issue_columns,
+              value=[],
+              interactive=True,
+          )
+        with gr.Column():
+          info_columns_select = gr.CheckboxGroup(
+              label="Info Columns",
+              choices=info_columns,
+              value=default_columns,
+              interactive=True,
+          )
+    
     with gr.Row():
         task_select = gr.Dropdown(
             label="Task",
@@ -110,42 +128,35 @@ def get_demo(leaderboard_tab):
             interactive=True,
         )
 
-    with gr.Row():
-        columns_select = gr.CheckboxGroup(
-            label="Show columns",
-            choices=column_names,
-            value=default_columns,
-            interactive=True,
-        )
-
     with gr.Row():
         leaderboard_df = gr.DataFrame(display_df, datatype=types, interactive=False)
 
-    def update_leaderboard_records(model_id, dataset_id, columns, task):
+    def update_leaderboard_records(model_id, dataset_id, issue_columns, info_columns, task):
         global update_time
         if datetime.datetime.now() - update_time < datetime.timedelta(minutes=10):
             return gr.update()
         update_time = datetime.datetime.now()
         logger.info("Updating leaderboard records")
         leaderboard.records = get_records_from_dataset_repo(leaderboard.LEADERBOARD)
-        return filter_table(model_id, dataset_id, columns, task)
+        return filter_table(model_id, dataset_id, issue_columns, info_columns, task)
 
     leaderboard_tab.select(
         fn=update_leaderboard_records, 
-        inputs=[model_select, dataset_select, columns_select, task_select], 
+        inputs=[model_select, dataset_select, issue_columns_select, info_columns_select, task_select], 
         outputs=[leaderboard_df])
 
     @gr.on(
         triggers=[
             model_select.change,
             dataset_select.change,
-            columns_select.change,
+            issue_columns_select.change,
+            info_columns_select.change,
             task_select.change,
         ],
-        inputs=[model_select, dataset_select, columns_select, task_select],
+        inputs=[model_select, dataset_select, issue_columns_select, info_columns_select, task_select],
         outputs=[leaderboard_df],
     )
-    def filter_table(model_id, dataset_id, columns, task):
+    def filter_table(model_id, dataset_id, issue_columns, info_columns, task):
         logger.info("Filtering leaderboard records")
         records = leaderboard.records
         # filter the table based on task
@@ -156,8 +167,9 @@ def get_demo(leaderboard_tab):
         if dataset_id and dataset_id != "Any":
             df = df[(df["dataset_id"] == dataset_id)]
 
-        # filter the table based on the columns
-        df = df[columns]
+        # filter the table based on the columns 
+        issue_columns.sort()
+        df = df[["model_id"] + info_columns + issue_columns]
         types = get_types(df)
         display_df = get_display_df(df)
         return gr.update(value=display_df, datatype=types, interactive=False)

app_text_classification.py

@@ -6,6 +6,7 @@ from io_utils import read_scanners, write_scanners
 from text_classification_ui_helpers import (
     get_related_datasets_from_leaderboard,
     align_columns_and_show_prediction,
+    get_dataset_splits,
     check_dataset,
     show_hf_token_info,
     precheck_model_ds_enable_example_btn,
@@ -70,7 +71,7 @@ def get_demo():
     with gr.Row():
         example_input = gr.HTML(visible=False)
     with gr.Row():
-        example_prediction = gr.Label(label="Model Prediction Sample", visible=False)
+        example_prediction = gr.Label(label="Model Sample Prediction", visible=False)
 
     with gr.Row():
         with gr.Accordion(
@@ -94,7 +95,7 @@ def get_demo():
 
         run_inference = gr.Checkbox(value=True, label="Run with Inference API")
         inference_token = gr.Textbox(
-            placeholder="hf-xxxxxxxxxxxxxxxxxxxx",
+            placeholder="hf_xxxxxxxxxxxxxxxxxxxx",
             value="",
             label="HF Token for Inference API",
             visible=True,
@@ -109,7 +110,7 @@ def get_demo():
         )
 
     with gr.Accordion(label="Scanner Advance Config (optional)", open=False):
-        scanners = gr.CheckboxGroup(label="Scan Settings", visible=True)
+        scanners = gr.CheckboxGroup(visible=True)
 
         @gr.on(triggers=[uid_label.change], inputs=[uid_label], outputs=[scanners])
         def get_scanners(uid):
@@ -146,19 +147,17 @@ def get_demo():
         fn=get_related_datasets_from_leaderboard,
         inputs=[model_id_input],
         outputs=[dataset_id_input],
-    ).then(
-        fn=check_dataset, 
-        inputs=[dataset_id_input], 
-        outputs=[dataset_config_input, dataset_split_input, loading_status]
     )
     
     gr.on(
-        triggers=[dataset_id_input.change],
+        triggers=[dataset_id_input.input, dataset_id_input.select],
         fn=check_dataset,
         inputs=[dataset_id_input],
         outputs=[dataset_config_input, dataset_split_input, loading_status]
     )
 
+    dataset_config_input.change(fn=get_dataset_splits, inputs=[dataset_id_input, dataset_config_input], outputs=[dataset_split_input])
+
     gr.on(
         triggers=[model_id_input.change, dataset_id_input.change, dataset_config_input.change],
         fn=empty_column_mapping,

text_classification.py

@@ -393,4 +393,21 @@ def check_hf_token_validity(hf_token):
     response = requests.get(AUTH_CHECK_URL, headers=headers)
     if response.status_code != 200:
         return False
-    return True
+    return True
+
+def get_dataset_info_from_server(dataset_id):
+    url = "https://datasets-server.huggingface.co/splits?dataset=" + dataset_id
+    response = requests.get(url)
+    if response.status_code != 200:
+        return None
+    return response.json()
+
+def get_dataset_splits(dataset_id, dataset_config):
+    dataset_info = get_dataset_info_from_server(dataset_id)
+    if dataset_info is None:
+        return None
+    try:
+        splits = dataset_info["splits"]
+        return [split["split"] for split in splits if split["config"] == dataset_config]
+    except Exception:
+        return None

text_classification_ui_helpers.py

@@ -26,7 +26,7 @@ from wordings import (
     NOT_TEXT_CLASSIFICATION_MODEL_RAW,
     UNMATCHED_MODEL_DATASET_STYLED_ERROR,
     CHECK_LOG_SECTION_RAW,
-    get_styled_input,
+    VALIDATED_MODEL_DATASET_STYLED,
     get_dataset_fetch_error_raw,
 )
 import os
@@ -44,13 +44,20 @@ def get_related_datasets_from_leaderboard(model_id):
     datasets_unique = list(model_records["dataset_id"].unique())
 
     if len(datasets_unique) == 0:
-        return gr.update(choices=[], value="")
+        return gr.update(choices=[])
     
-    return gr.update(choices=datasets_unique, value="")
+    return gr.update(choices=datasets_unique)
 
 
 logger = logging.getLogger(__file__)
 
+def get_dataset_splits(dataset_id, dataset_config):
+    try:
+        splits = datasets.get_dataset_split_names(dataset_id, dataset_config, trust_remote_code=True)
+        return gr.update(choices=splits, value=splits[0], visible=True)
+    except Exception as e:
+        logger.warn(f"Check your dataset {dataset_id} and config {dataset_config}: {e}")
+        return gr.update(visible=False)
 
 def check_dataset(dataset_id):
     logger.info(f"Loading {dataset_id}")
@@ -62,9 +69,7 @@ def check_dataset(dataset_id):
                 gr.update(visible=False),
                 ""
             )
-        splits = datasets.get_dataset_split_names(
-                        dataset_id, configs[0], trust_remote_code=True
-                    )
+        splits = datasets.get_dataset_split_names(dataset_id, configs[0], trust_remote_code=True)
         return (
             gr.update(choices=configs, value=configs[0], visible=True),
             gr.update(choices=splits, value=splits[0], visible=True),
@@ -169,21 +174,24 @@ def list_labels_and_features_from_dataset(ds_labels, ds_features, model_labels,
 
 def precheck_model_ds_enable_example_btn(
     model_id, dataset_id, dataset_config, dataset_split
-):
+):  
     model_id = strip_model_id_from_url(model_id)
     model_task = check_model_task(model_id)
     preload_hf_inference_api(model_id)
+
     if model_task is None or model_task != "text-classification":
         gr.Warning(NOT_TEXT_CLASSIFICATION_MODEL_RAW)
-        return (gr.update(), gr.update(),"")
+        return (gr.update(interactive=False), gr.update(visible=False),"")
 
     if dataset_config is None or dataset_split is None or len(dataset_config) == 0:
-        return (gr.update(), gr.update(), "")
+        return (gr.update(interactive=False), gr.update(visible=False), "")
     
     try:
         ds = datasets.load_dataset(dataset_id, dataset_config, trust_remote_code=True)
         df: pd.DataFrame = ds[dataset_split].to_pandas().head(5)
         ds_labels, ds_features = get_labels_and_features_from_dataset(ds[dataset_split])
+        if model_id == "" or model_id is None:
+          return (gr.update(interactive=False), gr.update(value=df, visible=True), "")
 
         if not isinstance(ds_labels, list) or not isinstance(ds_features, list):
             gr.Warning(CHECK_CONFIG_OR_SPLIT_RAW)
@@ -193,7 +201,7 @@ def precheck_model_ds_enable_example_btn(
     except Exception as e:
         # Config or split wrong
         logger.warn(f"Check your dataset {dataset_id} and config {dataset_config} on split {dataset_split}: {e}")
-        return (gr.update(interactive=False), gr.update(value=pd.DataFrame(), visible=False), "")
+        return (gr.update(interactive=False), gr.update(visible=False), "")
 
 
 def align_columns_and_show_prediction(
@@ -298,8 +306,8 @@ def align_columns_and_show_prediction(
         )
 
     return (
-        gr.update(value=get_styled_input(prediction_input), visible=True),
-        gr.update(value=prediction_response, visible=True),
+        gr.update(value=VALIDATED_MODEL_DATASET_STYLED, visible=True),
+        gr.update(value=prediction_response, label=prediction_input, visible=True),
         gr.update(visible=True, open=False),
         gr.update(interactive=(run_inference and inference_token != "")),
         "",

wordings.py

@@ -8,7 +8,7 @@ CONFIRM_MAPPING_DETAILS_MD = """
                             <h1 style="text-align: center;">
                             Confirm Pre-processing Details
                             </h1>
-                            Make sure the output variable's labels and the input variable's name are accurately mapped across both the dataset and the model.
+                            Make sure the output variable's labels and the input variable's name are accurately mapped across both the dataset and the model. You can select the output variable's labels from the dropdowns below.
                             """
 CONFIRM_MAPPING_DETAILS_FAIL_MD = """
                             <h1 style="text-align: center;">
@@ -38,7 +38,7 @@ PREDICTION_SAMPLE_MD = """
 
 MAPPING_STYLED_ERROR_WARNING = """
                         <h3 style="text-align: center;color: orange; background-color: #fff0f3; border-radius: 8px; padding: 10px; ">
-                        ⚠️ We're unable to automatically map the input variable's name and output variable's labels of your dataset with the model's. <b>Please manually check the mapping below.</b>
+                        ⚠️ We're unable to automatically map the input variable's name and output variable's labels of your dataset with the model's. Please manually check the mapping below.
                         </h3>
                         """
 
@@ -57,7 +57,7 @@ USE_INFERENCE_API_TIP = """
                 <a href="https://huggingface.co/docs/api-inference/detailed_parameters#text-classification-task">
                     Hugging Face Inference API
                 </a>
-                . Please input your <a href="https://huggingface.co/settings/tokens">Hugging Face token</a> to do so.
+                . Please input your <a href="https://huggingface.co/settings/tokens">Hugging Face token</a> to do so. You can find it <a href="https://huggingface.co/settings/tokens">here</a>.
             """
 
 HF_TOKEN_INVALID_STYLED= """
@@ -66,10 +66,10 @@ HF_TOKEN_INVALID_STYLED= """
                 </p>
                 """
 
+VALIDATED_MODEL_DATASET_STYLED = """
+            <h3 style="text-align: center;color: #4ca154; background-color: #e2fbe8; border-radius: 8px; padding: 10px; ">
+            Your model and dataset have been validated!
+            </h3>"""
+
 def get_dataset_fetch_error_raw(error):
     return f"""Sorry you cannot use this dataset because {error}. Contact HF team to support this dataset."""
-
-def get_styled_input(input):
-    return f"""<h3 style="text-align: center;color: #4ca154; background-color: #e2fbe8; border-radius: 8px; padding: 10px; ">
-            Your model and dataset have been validated! <br /> Sample input: {input}
-            </h3>"""