Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import pandas as pd | |
| title = """ | |
| # hmLeaderboard: Space for tracking and ranking models on Historical NER Datasets | |
|  | |
| """ | |
| description = """ | |
| ## Models | |
| At the moment the following backbone LMs are supported: | |
| * hmBERT: [Historical Multilingual Language Models for Named Entity Recognition](https://huggingface.co/hmbert). | |
| * hmTEAMS: [Historical Multilingual TEAMS Models](https://huggingface.co/hmteams). | |
| * hmByT5: [Historical Multilingual and Monolingual ByT5 Models](https://huggingface.co/hmbyt5) | |
| ## Datasets | |
| We test our pretrained language models on various datasets from HIPE-2020, HIPE-2022 and Europeana. The following table | |
| shows an overview of used datasets. | |
| | Language | Datasets | | |
| |----------|------------------------------------------------------------------| | |
| | English | [AjMC] - [TopRes19th] | | |
| | German | [AjMC] - [NewsEye] - [HIPE-2020] | | |
| | French | [AjMC] - [ICDAR-Europeana] - [LeTemps] - [NewsEye] - [HIPE-2020] | | |
| | Finnish | [NewsEye] | | |
| | Swedish | [NewsEye] | | |
| | Dutch | [ICDAR-Europeana] | | |
| [AjMC]: https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-ajmc.md | |
| [NewsEye]: https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-newseye.md | |
| [TopRes19th]: https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-topres19th.md | |
| [ICDAR-Europeana]: https://github.com/stefan-it/historic-domain-adaptation-icdar | |
| [LeTemps]: https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-letemps.md | |
| [HIPE-2020]: https://github.com/hipe-eval/HIPE-2022-data/blob/main/documentation/README-hipe2020.md | |
| ## Results | |
| """ | |
| 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 | |
| dataset_to_description_mapping = { | |
| "AjMC": "#### AjMC\nThe AjMC dataset consists of NE-annotated historical commentaries in the field of Classics, and was created in the context of the [Ajax MultiCommentary](https://mromanello.github.io/ajax-multi-commentary/) project.\n\nThe following NEs were annotated: `pers`, `work`, `loc`, `object`, `date` and `scope`.", | |
| "NewsEye": "#### NewsEye\nThe NewsEye dataset is comprised of diachronic historical newspaper material published between 1850 and 1950 in French, German, Finnish, and Swedish. More information can be found [here](https://dl.acm.org/doi/abs/10.1145/3404835.3463255).\n\nThe following NEs were annotated: `PER`, `LOC`, `ORG` and `HumanProd`.", | |
| "ICDAR": "#### ICDAR\nThe ICDAR-Europeana NER Dataset is a preprocessed variant of the [Europeana NER Corpora](https://github.com/EuropeanaNewspapers/ner-corpora) for Dutch and French.\n\nThe following NEs were annotated: `PER`, `LOC` and `ORG`.", | |
| "LeTemps": "#### LeTemps\nThe LeTemps dataset consists of NE-annotated historical French newspaper articles from mid-19C to mid 20C.\n\nThe following NEs were annotated: `loc`, `org` and `pers`.", | |
| "TopRes19th": "#### TopRes19th\nThe TopRes19th dataset consists of NE-annotated historical English newspaper articles from 19C.\n\nThe following NEs were annotated: `BUILDING`, `LOC` and `STREET`.", | |
| "HIPE-2020": "#### HIPE-2020\nThe HIPE-2020 dataset is comprised of newspapers from mid 19C to mid 20C. For information can be found [here](https://dl.acm.org/doi/abs/10.1007/978-3-030-58219-7_21).\n\nThe following NEs were annotated: `loc`, `org`, `pers`, `prod`, `time` and `comp`.", | |
| } | |
| configuration_to_description_mapping = { | |
| "Best Configuration": "The best hyper-parameter configuration for each model is used and average F1-score over runs with different seeds is reported here:", | |
| "Best Model": "The best hyper-parameter configuration for each model is used, the model with highest F1-score is chosen and its performance is reported here:" | |
| } | |
| with gr.Blocks() as demo: | |
| gr.Markdown(title) | |
| gr.Markdown(description) | |
| with gr.Tab("Overview"): | |
| gr.Markdown("### Best Configuration") | |
| gr.Markdown(configuration_to_description_mapping["Best Configuration"]) | |
| df_result = perform_evaluation_for_datasets("Best Configuration", dataset_names) | |
| gr.Dataframe(value=df_result) | |
| gr.Markdown("### Best Model") | |
| gr.Markdown(configuration_to_description_mapping["Best Model"]) | |
| df_result = perform_evaluation_for_datasets("Best Model", dataset_names) | |
| gr.Dataframe(value=df_result) | |
| for dataset_name, dataset_description in dataset_to_description_mapping.items(): | |
| with gr.Tab(dataset_name): | |
| selected_datasets = [ds for ds in dataset_names if dataset_name.lower() in ds.lower()] | |
| gr.Markdown(dataset_description) | |
| for config in ["Best Configuration", "Best Model"]: | |
| gr.Markdown(f"##### Results for {config}") | |
| gr.Markdown(configuration_to_description_mapping[config]) | |
| df_result = perform_evaluation_for_datasets(config, selected_datasets) | |
| 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() | |