Fix for flattened raw config

app.py

@@ -10,7 +10,7 @@ from run_jobs import start_process_run_job, stop_thread
 try:
     with gr.Blocks(theme=gr.themes.Soft(primary_hue="green")) as demo:
         with gr.Tab("Text Classification"):
-            get_demo_text_classification(demo)
+            get_demo_text_classification()
         with gr.Tab("Leaderboard"):
             get_demo_leaderboard()
         with gr.Tab("Logs(Debug)"):

app_text_classification.py

@@ -12,7 +12,7 @@ from text_classification_ui_helpers import (check_dataset_and_get_config,
                                             write_column_mapping_to_config)
 from wordings import CONFIRM_MAPPING_DETAILS_MD, INTRODUCTION_MD
 
-MAX_LABELS = 20
+MAX_LABELS = 40
 MAX_FEATURES = 20
 
 EXAMPLE_MODEL_ID = "cardiffnlp/twitter-roberta-base-sentiment-latest"
@@ -20,7 +20,7 @@ EXAMPLE_DATA_ID = "tweet_eval"
 CONFIG_PATH = "./config.yaml"
 
 
-def get_demo(demo):
+def get_demo():
     with gr.Row():
         gr.Markdown(INTRODUCTION_MD)
         uid_label = gr.Textbox(
@@ -55,9 +55,11 @@ def get_demo(demo):
             column_mappings = []
             with gr.Row():
                 with gr.Column():
+                    gr.Markdown("# Label Mapping")
                     for _ in range(MAX_LABELS):
                         column_mappings.append(gr.Dropdown(visible=False))
                 with gr.Column():
+                    gr.Markdown("# Feature Mapping")
                     for _ in range(MAX_LABELS, MAX_LABELS + MAX_FEATURES):
                         column_mappings.append(gr.Dropdown(visible=False))
 
@@ -138,9 +140,6 @@ def get_demo(demo):
         triggers=[label.change for label in column_mappings],
         fn=write_column_mapping_to_config,
         inputs=[
-            dataset_id_input,
-            dataset_config_input,
-            dataset_split_input,
             uid_label,
             *column_mappings,
         ],
@@ -151,9 +150,6 @@ def get_demo(demo):
         triggers=[label.input for label in column_mappings],
         fn=write_column_mapping_to_config,
         inputs=[
-            dataset_id_input,
-            dataset_config_input,
-            dataset_split_input,
             uid_label,
             *column_mappings,
         ],
@@ -172,6 +168,7 @@ def get_demo(demo):
             dataset_id_input,
             dataset_config_input,
             dataset_split_input,
+            uid_label,
         ],
         outputs=[
             example_input,

io_utils.py

@@ -76,7 +76,6 @@ def read_column_mapping(uid):
         config = yaml.load(f, Loader=yaml.FullLoader)
         if config:
             column_mapping = config.get("column_mapping", dict())
-    f.close()
     return column_mapping
 
 
@@ -84,7 +83,6 @@ def read_column_mapping(uid):
 def write_column_mapping(mapping, uid):
     with open(get_yaml_path(uid), "r") as f:
         config = yaml.load(f, Loader=yaml.FullLoader)
-    f.close()
 
     if config is None:
         return
@@ -92,10 +90,9 @@ def write_column_mapping(mapping, uid):
         del config["column_mapping"]
     else:
         config["column_mapping"] = mapping
-
     with open(get_yaml_path(uid), "w") as f:
-        yaml.dump(config, f, Dumper=Dumper)
-    f.close()
+        # yaml Dumper will by default sort the keys
+        yaml.dump(config, f, Dumper=Dumper, sort_keys=False)
 
 
 # convert column mapping dataframe to json

text_classification.py

@@ -16,15 +16,16 @@ def get_labels_and_features_from_dataset(dataset_id, dataset_config, split):
         ds = datasets.load_dataset(dataset_id, dataset_config)[split]
         dataset_features = ds.features
         label_keys = [i for i in dataset_features.keys() if i.startswith('label')]
-        if len(label_keys) == 0:
-            raise ValueError("Dataset does not have label column")
+        if len(label_keys) == 0: # no labels found
+            # return everything for post processing
+            return list(dataset_features.keys()), list(dataset_features.keys())
         if not isinstance(dataset_features[label_keys[0]], datasets.ClassLabel):
             if hasattr(dataset_features[label_keys[0]], 'feature'):
                 label_feat = dataset_features[label_keys[0]].feature
                 labels = label_feat.names
         else:
             labels = [dataset_features[label_keys[0]].names]
-        features = [f for f in dataset_features.keys() if f != "label"]
+        features = [f for f in dataset_features.keys() if not f.startswith("label")]
         return labels, features
     except Exception as e:
         logging.warning(

text_classification_ui_helpers.py

@@ -18,7 +18,7 @@ from wordings import (CHECK_CONFIG_OR_SPLIT_RAW,
                       CONFIRM_MAPPING_DETAILS_FAIL_RAW,
                       MAPPING_STYLED_ERROR_WARNING)
 
-MAX_LABELS = 20
+MAX_LABELS = 40
 MAX_FEATURES = 20
 
 HF_REPO_ID = "HF_REPO_ID"
@@ -68,46 +68,62 @@ def deselect_run_inference(run_local):
 
 
 def write_column_mapping_to_config(
-    dataset_id, dataset_config, dataset_split, uid, *labels
+    uid, *labels
 ):
     # TODO: Substitute 'text' with more features for zero-shot
     # we are not using ds features because we only support "text" for now
-    ds_labels, _ = get_labels_and_features_from_dataset(
-        dataset_id, dataset_config, dataset_split
-    )
+    all_mappings = read_column_mapping(uid)
+
     if labels is None:
         return
+    all_mappings = export_mappings(all_mappings, "labels", None, labels[:MAX_LABELS])
+    all_mappings = export_mappings(all_mappings, "features", ["text"], labels[MAX_LABELS : (MAX_LABELS + MAX_FEATURES)])
 
-    all_mappings = dict()
-
-    if "labels" not in all_mappings.keys():
-        all_mappings["labels"] = dict()
-    for i, label in enumerate(labels[:MAX_LABELS]):
-        if label:
-            all_mappings["labels"][label] = ds_labels[i % len(ds_labels)]
-    if "features" not in all_mappings.keys():
-        all_mappings["features"] = dict()
-    for _, feat in enumerate(labels[MAX_LABELS : (MAX_LABELS + MAX_FEATURES)]):
-        if feat:
-            # TODO: Substitute 'text' with more features for zero-shot
-            all_mappings["features"]["text"] = feat
     write_column_mapping(all_mappings, uid)
 
-
-def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label):
+def export_mappings(all_mappings, key, subkeys, values):
+    if key not in all_mappings.keys():
+        all_mappings[key] = dict()
+    if subkeys is None:
+        subkeys = list(all_mappings[key].keys())
+
+    if not subkeys:
+        logging.debug(f"subkeys is empty for {key}")
+        return all_mappings
+    
+    for i, subkey in enumerate(subkeys):
+        if subkey:
+            all_mappings[key][subkey] = values[i % len(values)]
+    return all_mappings
+
+def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label, uid):
     model_labels = list(model_id2label.values())
-    len_model_labels = len(model_labels)
+    all_mappings = read_column_mapping(uid)
+    # For flattened raw datasets with no labels
+    # check if there are shared labels between model and dataset
+    shared_labels = set(model_labels).intersection(set(ds_labels))
+    if shared_labels:
+        ds_labels = list(shared_labels)
+    if len(ds_labels) > MAX_LABELS:
+        ds_labels = ds_labels[:MAX_LABELS]
+        gr.Warning(f"The number of labels is truncated to length {MAX_LABELS}")
+
+    ds_labels.sort()
+    model_labels.sort()
+
     lables = [
         gr.Dropdown(
             label=f"{label}",
             choices=model_labels,
-            value=model_id2label[i % len_model_labels],
+            value=model_id2label[i % len(model_labels)],
             interactive=True,
             visible=True,
         )
-        for i, label in enumerate(ds_labels[:MAX_LABELS])
+        for i, label in enumerate(ds_labels)
     ]
     lables += [gr.Dropdown(visible=False) for _ in range(MAX_LABELS - len(lables))]
+    all_mappings = export_mappings(all_mappings, "labels", ds_labels, model_labels)
+
     # TODO: Substitute 'text' with more features for zero-shot
     features = [
         gr.Dropdown(
@@ -122,11 +138,14 @@ def list_labels_and_features_from_dataset(ds_labels, ds_features, model_id2label
     features += [
         gr.Dropdown(visible=False) for _ in range(MAX_FEATURES - len(features))
     ]
+    all_mappings = export_mappings(all_mappings, "features", ["text"], ds_features)
+    write_column_mapping(all_mappings, uid)
+
     return lables + features
 
 
 def check_model_and_show_prediction(
-    model_id, dataset_id, dataset_config, dataset_split
+    model_id, dataset_id, dataset_config, dataset_split, uid
 ):
     ppl = check_model(model_id)
     if ppl is None or not isinstance(ppl, TextClassificationPipeline):
@@ -168,6 +187,7 @@ def check_model_and_show_prediction(
         ds_labels,
         ds_features,
         model_id2label,
+        uid,
     )
 
     # when labels or features are not aligned