app.py

import gradio as gr
from transformers import pipeline
from utils import *
from datasets import load_dataset

pipe = pipeline(model="raminass/scotus-v10", top_k=13, padding=True, truncation=True)
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"]))

max_textboxes = 100


# https://www.gradio.app/guides/controlling-layout
def greet(opinion):
    chunks = chunk_data(remove_citations(opinion))["text"].to_list()
    result = average_text(chunks, pipe)
    k = len(chunks)
    wrt_boxes = []
    for i in range(k):
        wrt_boxes.append(gr.Textbox(chunks[i], visible=True))
        wrt_boxes.append(gr.Label(value=result[1][i], visible=True))
    return (
        [result[0]]
        + wrt_boxes
        + [gr.Textbox(visible=False), gr.Label(visible=False)] * (max_textboxes - k)
    )


def set_input(drop):
    return drop


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))

    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
        )
        textboxes.append(t)
        textboxes.append(par_level)

    drop.select(set_input, inputs=drop, outputs=[opinion])

    greet_btn.click(
        fn=greet,
        inputs=opinion,
        outputs=[op_level] + textboxes,
    )


if __name__ == "__main__":
    demo.launch()