Upload folder using huggingface_hub

app.py

@@ -8,16 +8,23 @@ all = load_dataset("raminass/full_opinions_1994_2020")
 df = pd.DataFrame(all["train"])
 choices = []
 for index, row in df[df.category == "per_curiam"].iterrows():
-    choices.append((f"""{row["case_name"]}""", row["text"]))
+    choices.append((f"""{row["case_name"]}""", [row["text"], row["year_filed"]]))
 
 max_textboxes = 100
 
 
 # https://www.gradio.app/guides/controlling-layout
-def greet(opinion):
+def greet(opinion, year):
+    judges_l = (
+        df[(df["year_filed"] == year) & (df["category"] != "per_curiam")]
+        .author_name.unique()
+        .tolist()
+    )
+
     chunks = chunk_data(remove_citations(opinion))["text"].to_list()
-    result = average_text(chunks, pipe)
+    result = average_text(chunks, pipe, judges_l)
     k = len(chunks)
+
     wrt_boxes = []
     for i in range(k):
         wrt_boxes.append(gr.Textbox(chunks[i], visible=True))
@@ -30,30 +37,33 @@ def greet(opinion):
 
 
 def set_input(drop):
-    return drop
+    return drop[0], drop[1]
 
 
 with gr.Blocks() as demo:
-    opinion = gr.Textbox(label="Opinion")
-    op_level = gr.outputs.Label(num_top_classes=13, label="Overall")
-
-    drop = gr.Dropdown(choices=sorted(choices))
+    with gr.Row():
+        with gr.Column():
+            opinion = gr.Textbox(label="Opinion")
+            year = gr.Slider(1994, 2020, label="Year")
+            drop = gr.Dropdown(choices=sorted(choices))
+            greet_btn = gr.Button("Predict")
+        op_level = gr.outputs.Label(num_top_classes=13, label="Overall")
 
-    greet_btn = gr.Button("Predict")
     textboxes = []
     for i in range(max_textboxes):
-        t = gr.Textbox(f"Textbox {i}", visible=False, label=f"Paragraph {i+1} Text")
-        par_level = gr.Label(
-            num_top_classes=5, label=f"Paragraph {i+1} Prediction", visible=False
-        )
+        with gr.Row():
+            t = gr.Textbox(f"Textbox {i}", visible=False, label=f"Paragraph {i+1} Text")
+            par_level = gr.Label(
+                num_top_classes=5, label=f"Paragraph {i+1} Prediction", visible=False
+            )
         textboxes.append(t)
         textboxes.append(par_level)
 
-    drop.select(set_input, inputs=drop, outputs=[opinion])
+    drop.select(set_input, inputs=drop, outputs=[opinion, year])
 
     greet_btn.click(
         fn=greet,
-        inputs=opinion,
+        inputs=[opinion, year],
         outputs=[op_level] + textboxes,
     )
 

utils/__init__.py

@@ -10,19 +10,33 @@ with open("utils/label2id.json", "r") as j:
     label2id = json.loads(j.read())
 
 
-def average_text(text, model):
-    # result = classifier(df_train[(df_train.case_name==case) & (df_train.category=='per_curiam')]['clean_text'].to_list())
+def normaliz_dict(d, target=1.0):
+    raw = sum(d.values())
+    factor = target / raw
+    return {key: value * factor for key, value in d.items()}
+
+
+def average_text(text, model, judges):
     result = model(text)
+    new_res = []
+    for d in result:
+        p = {}
+        for dicts in d:
+            if dicts["label"] in judges:
+                p[dicts["label"]] = round(dicts["score"], 2)
+        p = normaliz_dict(p)
+        new_res.append(p)
+
     pred = {}
-    for c in result:
-        for d in c:
-            if d["label"] not in pred:
-                pred[d["label"]] = [round(d["score"], 2)]
+    for c in new_res:
+        for k, v in c.items():
+            if k not in pred:
+                pred[k] = [round(v, 2)]
             else:
-                pred[d["label"]].append(round(d["score"], 2))
+                pred[k].append(round(v, 2))
     sumary = {k: round(sum(v) / len(v), 2) for k, v in pred.items()}
-    result = [{dct["label"]: round(dct["score"], 2) for dct in lst} for lst in result]
-    return dict(sorted(sumary.items(), key=lambda x: x[1], reverse=True)), result
+    sumary = normaliz_dict(sumary)
+    return dict(sorted(sumary.items(), key=lambda x: x[1], reverse=True)), new_res
 
 
 # def find_case_by_name(df, name):