app.py

import os
import random
import uuid
from datetime import datetime

import gradio as gr

from data_loader import load_data

HF_TOKEN = os.environ.get('HF_TOKEN')
HF_DATASET = os.environ.get('HF_DATASET')
N_QUESTIONS = 5

data, models = load_data()

n_samples = len(data)
n_models = len(models)

saver = gr.HuggingFaceDatasetSaver(HF_TOKEN, HF_DATASET, private=True)


def convert_diff_to_unified(diff):
    result = "\n".join(
        [
            f'--- {modified_file["old_path"]}\n'
            f'+++ {modified_file["new_path"]}\n'
            f'{modified_file["diff"]}'
            for modified_file in diff
        ]
    )

    return result


def get_diff2html_view(raw_diff):
    html = f"""
    <div style='width:100%; height:1400px; overflow:auto; position: relative'>
        <div id='diff-raw' hidden>{raw_diff}</div> 
        <div class="d2h-view-wrapper">
            <div id='diff-view'></div>
        </div>
    </div>
    """

    return html


def get_github_link_md(repo, hash):
    return f'[See the commit on Github](https://github.com/{repo}/commit/{hash})'


def update_commit_view(sample_ind):
    if sample_ind >= n_samples:
        return None

    record = data[sample_ind]

    diff_view = get_diff2html_view(convert_diff_to_unified(record['mods']))

    repo_val = record['repo']
    hash_val = record['hash']
    github_link_md = get_github_link_md(repo_val, hash_val)

    diff_loaded_timestamp = datetime.now().isoformat()

    models_shuffled = models[:]
    random.shuffle(models_shuffled)

    commit_messages = tuple(record[model] for model in models_shuffled)

    return (github_link_md, diff_view, repo_val, hash_val, diff_loaded_timestamp) + commit_messages + tuple(
        models_shuffled)


def reset_answers():
    return (None,) * (n_models * N_QUESTIONS)


def next_sample(current_sample_ind, shuffled_idx):
    if current_sample_ind == n_samples:
        return None

    current_sample_ind += 1
    updated_view = update_commit_view(shuffled_idx[current_sample_ind])
    return (current_sample_ind,) + updated_view + reset_answers()


with open("head.html") as head_file:
    head_html = head_file.read()

with gr.Blocks(theme=gr.themes.Soft(), head=head_html, css="style_overrides.css") as application:
    repo_val = gr.Textbox(interactive=False, label='repo', visible=False)
    hash_val = gr.Textbox(interactive=False, label='hash', visible=False)
    shuffled_idx_val = gr.JSON(visible=False)

    with gr.Row():
        current_sample_sld = gr.Slider(minimum=0, maximum=n_samples, step=1,
                                       value=0,
                                       interactive=False,
                                       label='sample_ind',
                                       info=f"Samples labeled/skipped (out of {n_samples})",
                                       show_label=False,
                                       container=False,
                                       scale=5)
        with gr.Column(scale=1):
            skip_btn = gr.Button("Skip the current sample")
    with gr.Row():
        with gr.Column(scale=2):
            github_link = gr.Markdown()
            diff_view = gr.HTML()
        with gr.Column(scale=1):
            commit_msg = []
            is_correct = []
            has_what = []
            has_why = []
            is_not_verbose = []
            has_headline = []
            easy_to_read = []
            overall_rating = []
            comments = []
            model_name = []

            SCALE = list(range(1, 6))

            for model_ind in range(n_models):
                with gr.Tab(f"Variant #{model_ind + 1}"):
                    commit_msg.append(gr.TextArea(label="Commit message (can be scrollable)",
                                                  interactive=False,
                                                  ))
                    gr.Markdown("## Please, rate your level of agreement with each statement\n"
                                "\n"
                                "*1 - strongly disagree, 2 - disagree, 3 - not sure, 4 - agree, 5 - strongly agree*")

                    is_correct.append(gr.Radio(
                        info='The information provided in the commit message is consistent with the code changes.',
                        label=f'is_correct_{model_ind}',
                        show_label=False,
                        choices=SCALE,
                        interactive=True))

                    # has_what.append(gr.Radio(
                    #     info='The commit message answers the question of WHAT changes have been made.',
                    #     label=f'has_what_{model_ind}',
                    #     show_label=False,
                    #     choices=SCALE,
                    #     interactive=True))
                    #
                    # has_why.append(gr.Radio(
                    #     info='The commit message answers the question of WHY these changes have been made.',
                    #     label=f'has_why_{model_ind}',
                    #     show_label=False,
                    #     choices=SCALE,
                    #     interactive=True))

                    is_not_verbose.append(gr.Radio(
                        info='The commit message can be substantially shortened without loss of important information.',
                        label=f'is_not_verbose_{model_ind}',
                        show_label=False,
                        choices=SCALE,
                        interactive=True))

                    # has_headline.append(gr.Radio(
                    #     info='The commit message includes a short headline that provides a good overview of the '
                    #          'changes.',
                    #     label=f'has_headline_{model_ind}',
                    #     show_label=False,
                    #     choices=SCALE,
                    #     interactive=True))

                    easy_to_read.append(gr.Radio(
                        info='The commit message is easy to read and to understand.',
                        label=f'easy_to_read_{model_ind}',
                        show_label=False,
                        choices=SCALE,
                        interactive=True))

                    overall_rating.append(gr.Radio(
                        info='Please, describe your overall impression of the commit message (1 - very bad, 5 - very '
                             'good)',
                        label=f'overall_rating_{model_ind}',
                        show_label=False,
                        choices=SCALE,
                        interactive=True))

                    comments.append(gr.Textbox(
                        info='Additional comments on the commit message',
                        label=f'comments_{model_ind}',
                        show_label=False,
                        interactive=True))

                    model_name.append(gr.Textbox(interactive=False, label=f'model_{model_ind}', visible=False))

            submit_btn = gr.Button("Submit all and continue")
            session_val = gr.Textbox(info='Session', interactive=False, container=True, show_label=False,
                                     label='session')

            with gr.Row(visible=False):
                sample_loaded_timestamp = gr.Textbox(info="Sample loaded", label='loaded_ts', interactive=False,
                                                     container=True, show_label=False)
                sample_submitted_timestamp = gr.Textbox(info="Current time",
                                                        interactive=False, container=True, show_label=False,
                                                        value=lambda: datetime.now().isoformat(), every=1.0,
                                                        label='submitted_ts')

    commit_view = [
        github_link,
        diff_view,
        repo_val,
        hash_val,
        sample_loaded_timestamp,
        *commit_msg,
        *model_name,
    ]

    feedback_metadata = [
        session_val,
        repo_val,
        hash_val,
        sample_loaded_timestamp,
        sample_submitted_timestamp,
        *model_name
    ]

    feedback_answers = [
        *is_correct,
        # *has_what,
        # *has_why,
        *is_not_verbose,
        # *has_headline,
        *easy_to_read,
        *overall_rating,
        *comments
    ]

    assert len(feedback_answers) == N_QUESTIONS * n_models

    saver.setup([current_sample_sld] + feedback_metadata + feedback_answers, "feedback")

    skip_btn.click(next_sample, inputs=[current_sample_sld, shuffled_idx_val],
                   outputs=[current_sample_sld] + commit_view + feedback_answers)


    def submit(current_sample, shuffled_idx, *args):
        saver.flag((current_sample,) + args)
        return next_sample(current_sample, shuffled_idx)


    submit_btn.click(submit, inputs=[current_sample_sld, shuffled_idx_val] + feedback_metadata + feedback_answers,
                     outputs=[current_sample_sld] + commit_view + feedback_answers)


    def init_session(current_sample):
        session = str(uuid.uuid4())
        shuffled_idx = list(range(n_samples))
        random.shuffle(shuffled_idx)
        return (session, shuffled_idx) + update_commit_view(shuffled_idx[current_sample])


    application.load(init_session,
                     inputs=[current_sample_sld],
                     outputs=[session_val, shuffled_idx_val] + commit_view, )

application.launch()