app.py

import gradio as gr
import arxiv
import json
from model import predict_from_text

with open("./data/arxiv-label-dict.json", "r") as file:
    subject_dict = json.loads(file.read())


def parse_id(input_id):
    ## Grab article title and true categories from arXiv
    search = arxiv.Search(id_list=[input_id], max_results=1)
    result = next(search.results())
    raw_categories = result.categories
    title = result.title
    abstract = result.summary
    subject_tags = ", ".join(
        sorted(
            [subject_dict[tag] for tag in raw_categories if tag in subject_dict.keys()]
        )
    )

    return (title, subject_tags, abstract)


# def parse_title(input_title):
#     query_title = input_title.replace(" ", "\%20")
#     search = arxiv.Search(
#         query=f"ti:\%22{query_title}\%22",
#         sort_by=arxiv.SortCriterion.Relevance,
#         sort_order=arxiv.SortOrder.Descending,
#         max_results=1,
#     )
#     result = next(search.results())
#     raw_categories = result.categories
#     title = result.title

#     with open("./data/arxiv-label-dict.json", "r") as file:
#         subject_dict = json.loads(file.read())

#     subject_tags = ", ".join(
#         sorted(
#             [subject_dict[tag] for tag in raw_categories if tag in subject_dict.keys()]
#         )
#     )

#     return (title, subject_tags)


def outputs_from_id(input_id, threshold_probability):
    title, true_tags, abstract = parse_id(input_id)
    predicted_tags = predict_from_text(title, threshold_probability)
    return title, predicted_tags, true_tags, abstract


# def outputs_from_title(input_title, threshold_probability):
#     title, true_tags = parse_title(input_title)
#     predicted_tags = predict_from_text(title, threshold_probability)

#     return title, predicted_tags, true_tags


with gr.Blocks() as demo:
    gr.Markdown(
        """# <center> Math Subject Classifier 
                ## [Search](https://arxiv.org) for a math article and input its ID number below. </center>"""
    )
    with gr.Row():
        id_input = gr.Textbox(label="arXiv ID:", placeholder="XXXX.XXXXX")
        id_title = gr.Textbox(label="Title of Input Article:")
        id_predict = gr.Textbox(label="Predicted Subject Tags:")
        id_true = gr.Textbox(label="Actual Subject Tags:")
    threshold_probability = gr.Slider(
        label="Minimum Confidence For Tag Prediction:", value=0.5, minimum=0, maximum=1
    )
    id_button = gr.Button("Get Predicted Subject Tags")
    gr.Examples(
        label="Try These Example Articles:",
        examples=[
            "1709.07343",
            "2107.05105",
            "1910.06441",
            "2210.09246",
            "2111.03188",
            "1811.07007",
            "2303.15347",
            "2210.04580",
            "1909.06032",
            "2107.13138",
        ],
        inputs=id_input,
    )
    gr.Markdown("### Article Abstract:")
    article_abstract = gr.HTML()

    # with gr.Tab("Predict by title"):
    #     with gr.Row():
    #         title_input = gr.Textbox(label="Input title")
    #         title_title = gr.Textbox(label="Title of closest match")
    #         title_predict = gr.Textbox(label="Predicted tags")
    #         title_true = gr.Textbox(label="True tags")
    #     title_button = gr.Button("Predict")
    #     gr.Examples(
    #         examples=[
    #             "Attention is all you need",
    #             "Etale cohomology of diamonds",
    #             "Stochastic Kahler geometry from random zeros to random metrics",
    #             "Scaling asymptotics for Szego kernels on Grauert tubes",
    #             "The Wave Trace and Birkhoff Billiards",
    #         ],
    #         inputs=title_input,
    #     )

    id_button.click(
        outputs_from_id,
        inputs=[id_input, threshold_probability],
        outputs=[id_title, id_predict, id_true, article_abstract],
    )
    # title_button.click(
    #     outputs_from_title,
    #     inputs=[title_input, threshold_probability],
    #     outputs=[title_title, title_predict, title_true],
    # )

demo.launch(inbrowser=True)