app.py

import gradio as gr
import pandas as pd

mase = pd.read_csv("results/results_mase.csv")
datasets = mase.dataset.unique()
frameworks = mase.framework.unique()
mase.set_index(["dataset", "framework"], inplace=True)

data = {"Dataset": datasets}


def mean(data, framework):
    try:
        return f"{round(mase.loc[data, framework].metric_error.mean(),3)} +/- {round(mase.loc[data, framework].metric_error.std(),3)}"
    except KeyError:
        return "n/a"


for framework in frameworks:
    data.update({framework: [mean(dataset, framework) for dataset in datasets]})

df = pd.DataFrame(data=data)

with gr.Blocks() as demo:
    gr.Markdown(
        """
    # Time Series Forecasting Leaderboard
    
    This is a leaderboard of the [MASE](https://huggingface.co/spaces/evaluate-metric/mase) metric for time series forecasting problem on the different open datasets and models.

    The table is generated from the paper [AutoGluon–TimeSeries: AutoML for Probabilistic Time Series Forecasting](https://github.com/autogluon/autogluon) by Oleksandr Shchur, Caner Turkmen, Nick Erickson, Huibin Shen, Alexander Shirkov, Tony Hu, and Bernie Wang.

    ## MASE Metric
    """
    )
    gr.Dataframe(df)

if __name__ == "__main__":
    demo.launch()