import gradio as gr import pandas as pd title = """ # hmLeaderboard ![hmLeaderboard](https://huggingface.co/spaces/hmbench/hmLeaderboard/resolve/main/logo.png) """ description = """ ## Space for tracking and ranking models on Historic NER Datasets. At the moment the following models are supported: * hmBERT: [Historical Multilingual Language Models for Named Entity Recognition](https://huggingface.co/hmbert). * hmTEAMS: [Historic Multilingual TEAMS Models](https://huggingface.co/hmteams). """ footer = "Made from Bavarian Oberland with ❤️ and 🥨." model_selection_file_names = { "Best Configuration": "best_model_configurations.csv", "Best Model": "best_models.csv" } df_init = pd.read_csv(model_selection_file_names["Best Configuration"]) dataset_names = df_init.columns.values[1:].tolist() languages = list(set([dataset_name.split(" ")[0] for dataset_name in dataset_names])) def perform_evaluation_for_datasets(model_selection, selected_datasets): df = pd.read_csv(model_selection_file_names.get(model_selection)) selected_indices = [] for selected_dataset in selected_datasets: selected_indices.append(dataset_names.index(selected_dataset) + 1) mean_column = df.iloc[:, selected_indices].mean(axis=1).round(2) # Include column with column name result_df = df.iloc[:, [0] + selected_indices] result_df["Average"] = mean_column return result_df def perform_evaluation_for_languages(model_selection, selected_languages): df = pd.read_csv(model_selection_file_names.get(model_selection)) selected_indices = [] for selected_language in selected_languages: selected_language = selected_language.lower() found_indices = [i for i, column_name in enumerate(df.columns) if selected_language in column_name.lower()] for found_index in found_indices: selected_indices.append(found_index) mean_column = df.iloc[:, selected_indices].mean(axis=1).round(2) # Include column with column name result_df = df.iloc[:, [0] + selected_indices] result_df["Average"] = mean_column return result_df with gr.Blocks() as demo: gr.Markdown(title) gr.Markdown(description) with gr.Tab("Overview"): gr.Markdown("### Best Configuration\nThe best hyper-parameter configuration for each model is used and average F1-score over runs with different seeds is reported here:") df_result = perform_evaluation_for_datasets("Best Configuration", dataset_names) gr.Dataframe(value=df_result) gr.Markdown("### Best Model\nThe best hyper-parameter configuration for each model is used and the model with highest F1-score is used and its performance is reported here:") df_result = perform_evaluation_for_datasets("Best Model", dataset_names) gr.Dataframe(value=df_result) with gr.Tab("Filtering"): gr.Markdown("### Filtering\nSwiss-knife filtering for single datasets and languages is possible.") model_selection = gr.Radio(choices=["Best Configuration", "Best Model"], label="Model Selection", info="Defines if best configuration or best model should be used for evaluation. When 'Best Configuration' is used, the best hyper-parameter configuration is used and then averaged F1-score over all runs is calculated. When 'Best Model' is chosen, the best hyper-parameter configuration and model with highest F1-score on development dataset is used (best model).", value="Best Configuration") with gr.Tab("Dataset Selection"): datasets_selection = gr.CheckboxGroup( dataset_names, label="Datasets", info="Select datasets for evaluation" ) output_df = gr.Dataframe() evaluation_button = gr.Button("Evaluate") evaluation_button.click(fn=perform_evaluation_for_datasets, inputs=[model_selection, datasets_selection], outputs=output_df) with gr.Tab("Language Selection"): language_selection = gr.CheckboxGroup( languages, label="Languages", info="Select languages for evaluation" ) output_df = gr.Dataframe() evaluation_button = gr.Button("Evaluate") evaluation_button.click(fn=perform_evaluation_for_languages, inputs=[model_selection, language_selection], outputs=output_df) gr.Markdown(footer) demo.launch()