app.py

from datetime import datetime

import streamlit as st
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

from datasets import Dataset
from load_dataframe import get_data


def aggregated_data(df, aggregation_level="week"):

    st.write(f"Aggregated data by {aggregation_level}")

    # Create a column that indicates if a paper has any artifacts
    df['has_artifact'] = (df['num_models'] > 0) | (df['num_datasets'] > 0) | (df['num_spaces'] > 0)

    # Resample by week
    freq = 'W' if aggregation_level == "week" else 'ME'
    weekly_total_papers = df.resample(freq).size()
    weekly_papers_with_artifacts = df.resample(freq)['has_artifact'].sum()

    # Calculate the percentage of papers with artifacts
    percentage_papers_with_artifacts = (weekly_papers_with_artifacts / weekly_total_papers) * 100

    # Calculate the growth rate
    growth_rate = percentage_papers_with_artifacts.pct_change() * 100
    growth_rate = growth_rate.replace([float('inf'), float('-inf')], pd.NA).dropna()

    # Display the average growth rate as a big number
    average_growth_rate = growth_rate.mean()
    st.metric(label=f"{aggregation_level.capitalize()}ly Average Growth Rate", value=f"{average_growth_rate:.2f}%")

    # Create the plot
    plt.figure(figsize=(12, 6))
    plt.plot(percentage_papers_with_artifacts.index, percentage_papers_with_artifacts, marker='o', linestyle='-', color='b', label='Percentage of Papers with on least 1 Artifact')

    # Set the y-axis limits
    plt.ylim(0, 100)
    
    plt.xlabel(aggregation_level)
    plt.ylabel('Percentage')
    plt.title('Percentage of Papers with Artifacts (Models, Datasets, Spaces) Over Time')
    plt.legend()
    plt.grid(True)

    # Use Streamlit to display the plot
    st.pyplot(plt)


def show_data_editor(df: pd.DataFrame, key: str):
    edited_df = st.data_editor(df,
                hide_index=True,
                column_order=("reached_out", "reached_out_link", "paper_page", "title", "github", "num_models", "num_datasets", "num_spaces"),
                column_config={"github": st.column_config.LinkColumn(),
                                "paper_page": st.column_config.LinkColumn(),
                                "paper_page_with_title": st.column_config.LinkColumn(display_text=r'\|(.*)')},
                width=2000,
                key=key)
    
    # Check if the dataframe has been edited
    # TODO this is wrong
    # rather we should probably do a merge-join (overwriting the edited rows) and then save the new dataframe
    # if not edited_df.equals(df):
    #     save_data(edited_df)
    #     st.success("Changes saved successfully!")


def save_data(df: pd.DataFrame):
    # load as HF dataset
    dataset = Dataset.from_pandas(df)

    dataset.push_to_hub("nielsr/daily-papers-enriched")

    return


def display_data(df: pd.DataFrame):
    df['has_artifact'] = (df['num_models'] > 0) | (df['num_datasets'] > 0) | (df['num_spaces'] > 0)
    num_artifacts = df['has_artifact'].sum()
    percentage_of_at_least_one_artifact = num_artifacts / df.shape[0] if df.shape[0] > 0 else 0
    percentage_of_at_least_one_artifact = round(percentage_of_at_least_one_artifact * 100, 2)

    # add reached out and reached out link columns
    df['reached_out'] = [False for _ in range(df.shape[0])]
    df["reached_out_link"] = ["" for _ in range(df.shape[0])]

    st.markdown(f"""
    ## {percentage_of_at_least_one_artifact}% papers with at least one 🤗 artifact
    
    * Number of papers: {df.shape[0]}
    * Number of papers with a Github link: {df['github'].notnull().sum()}
    * Number of papers with at least one HF artifact: {num_artifacts}
    """)

    st.write("Papers with at least one artifact")
    show_data_editor(df[df['has_artifact']], key="papers_with_artifacts")

    st.write("Papers without artifacts")
    show_data_editor(df[~df['has_artifact']], key="papers_without_artifacts")
    
    st.write("Papers with a HF mention in README but no artifacts")
    show_data_editor(df[(df['hf_mention'] == 1) & (~df['has_artifact'])], key="papers_with_hf_mention_no_artifacts")


def main():
    st.title("Hugging Face Artifacts KPI Dashboard")

    # 2 tabs: one for daily data, one for weekly data
    st.sidebar.title("Navigation")
    selection = st.sidebar.selectbox("Go to", ["Daily/weekly/monthly data", "Aggregated data"])

    if selection == "Daily/weekly/monthly data":
        # Button to select day, month or week
        # Add streamlit selectbox.
        view_level = st.selectbox(label="View data per day, week or month", options=["day", "week", "month"])

        if view_level == "day":
            # get the latest dataframe
            df = get_data()
            
            # make a button to select the day, defaulting to today
            day = st.date_input("Select day", value="today", format="DD/MM/YYYY")
            # convert to the day of a Pandas Timestamp
            day = pd.Timestamp(day)

            filtered_df = df[df.index.date == day.date()]

            st.write(f"Showing data for {day.day_name()} {day.strftime('%d/%m/%Y')}")
            display_data(df=filtered_df)

        elif view_level == "week":
            # get the latest dataframe
            df = get_data()
            
            # make a button to select the week
            week_number = st.number_input("Select week", value=datetime.today().isocalendar()[1], min_value=1, max_value=52)
            
            # Extract week number from the index
            df['week'] = df.index.isocalendar().week

            # Filter the dataframe for the desired week number
            filtered_df = df[df['week'] == week_number]
            
            st.write(f"Showing data for week {week_number}")
            
            display_data(df=filtered_df)

        elif view_level == "month":
            # get the latest dataframe
            df = get_data()
            
            # make a button to select the month, defaulting to current month
            month_str = st.selectbox("Select month", options=["January", "February", "March", "April", "May", "June", "July", "August", "September", "October", "November", "December"])
            year_str = st.selectbox("Select year", options=["2024"])
            
            # Filter the dataframe for the desired week number
            month_map = {
                'January': 1, 'February': 2, 'March': 3, 'April': 4, 
                'May': 5, 'June': 6, 'July': 7, 'August': 8, 
                'September': 9, 'October': 10, 'November': 11, 'December': 12
            }
            
            # Convert month string to number
            month = month_map[month_str]
            year = int(year_str)
            filtered_df = df[(df.index.month == month) & (df.index.year == year)]
            
            st.write(f"Showing data for {month_str} {year_str}")
            
            display_data(df=filtered_df)

    elif selection == "Aggregated data":

        # get the latest dataframe
        df = get_data()

        aggregated_data(df)
        aggregated_data(df, aggregation_level="month")

    else:
        st.write("Error: selection not recognized")


if __name__ == "__main__":
    main()