Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import pandas as pd | |
| import numpy as np | |
| # Global variable to store the DataFrame | |
| df = None | |
| # Global variable to keep track of the current row index | |
| current_row = 0 | |
| def load_csv(file): | |
| global df | |
| global current_row | |
| # import the csv and set the data types to be int, string, string, string, string, string, string | |
| df = pd.read_csv(file.name, dtype={'id':int, 'hs': str, 'cs': str, 'topic': str, 'tone': str, 'isCSContextuallyRelevant': str, 'isToneMatch': str}) | |
| if 'suggestedTone' not in df.columns: | |
| df['suggestedTone'] = None | |
| current_row = 0 | |
| row_dict = df.iloc[current_row].to_dict() | |
| return row_dict['id'], row_dict['hs'], row_dict['cs'], row_dict['topic'], row_dict['tone'], row_dict['isCSContextuallyRelevant'], row_dict['isToneMatch'], row_dict['suggestedTone'] | |
| def annotate_row(isCSContextuallyRelevant, isToneMatch, suggestedTone): | |
| global df | |
| global current_row | |
| df.at[current_row, 'isCSContextuallyRelevant'] = isCSContextuallyRelevant | |
| df.at[current_row, 'isToneMatch'] = isToneMatch | |
| df.at[current_row, 'suggestedTone'] = suggestedTone | |
| if current_row < len(df) - 1: | |
| current_row += 1 | |
| else: | |
| current_row = 0 | |
| df.to_csv('annotated_data.csv', index=False) | |
| row_dict = df.iloc[current_row].to_dict() | |
| return row_dict['id'], row_dict['hs'], row_dict['cs'], row_dict['topic'], row_dict['tone'], row_dict['isCSContextuallyRelevant'], row_dict['isToneMatch'], row_dict['suggestedTone'], 'annotated_data.csv' | |
| def navigate(direction): | |
| global current_row | |
| if direction == "Previous": | |
| current_row = max(0, current_row - 1) | |
| elif direction == "Next": | |
| current_row = min(len(df) - 1, current_row + 1) | |
| elif direction == "First Unlabeled": | |
| unlabeled_row = df[df['isCSContextuallyRelevant'].isna()].index.min() | |
| if not np.isnan(unlabeled_row): | |
| current_row = int(unlabeled_row) | |
| row_dict = df.iloc[current_row].to_dict() | |
| return row_dict['id'], row_dict['hs'], row_dict['cs'], row_dict['topic'], row_dict['tone'], row_dict['isCSContextuallyRelevant'], row_dict['isToneMatch'], row_dict['suggestedTone'] | |
| with gr.Blocks(theme=gr.themes.Soft()) as annotator: | |
| gr.Markdown("## Data Annotation") | |
| with gr.Row(): | |
| gr.Markdown("### Upload CSV") | |
| file_upload = gr.File() | |
| btn_load = gr.Button("Load CSV") | |
| with gr.Row(): | |
| gr.Markdown("### Current Row") | |
| with gr.Row(): | |
| idx = gr.Number(label='Index') | |
| hs = gr.Textbox(label='HS') | |
| cs = gr.Textbox(label='CS') | |
| with gr.Row(): | |
| topic = gr.Textbox(label='Topic') | |
| tone = gr.Textbox(label='Tone') | |
| with gr.Row(): | |
| isCSContextuallyRelevant = gr.Radio(["1", "0"], label="Contextually Relevant?") | |
| isToneMatch = gr.Radio(["1", "0"], label="Tone Match?") | |
| suggestedTone = gr.Dropdown(['', 'empathy', 'refutal', 'first_person', 'warn_of_conseq', 'humor', 'other'], label='Suggested Tone', interactive=True) | |
| btn_annotate = gr.Button("Annotate") | |
| with gr.Row(): | |
| btn_previous = gr.Button("Previous") | |
| btn_next = gr.Button("Next") | |
| btn_first_unlabeled = gr.Button("First Unlabeled") | |
| with gr.Row(): | |
| gr.Markdown("### Annotated Data File Download") | |
| file_download = gr.File() | |
| btn_load.click(load_csv, inputs=[file_upload], outputs=[idx, hs, cs, topic, tone, isCSContextuallyRelevant, isToneMatch, suggestedTone]) | |
| btn_annotate.click(annotate_row, inputs=[isCSContextuallyRelevant, isToneMatch, suggestedTone], outputs=[idx, hs, cs, topic, tone, isCSContextuallyRelevant, isToneMatch, suggestedTone, file_download]) | |
| btn_previous.click(navigate, inputs=gr.Textbox("Previous", visible=False), outputs=[idx, hs, cs, topic, tone, isCSContextuallyRelevant, isToneMatch, suggestedTone]) | |
| btn_next.click(navigate, inputs=gr.Textbox("Next", visible=False), outputs=[idx, hs, cs, topic, tone, isCSContextuallyRelevant, isToneMatch, suggestedTone]) | |
| btn_first_unlabeled.click(navigate, inputs=gr.Textbox("First Unlabeled", visible=False), outputs=[idx, hs, cs, topic, tone, isCSContextuallyRelevant, isToneMatch, suggestedTone]) | |
| annotator.launch() |