app.py

from datetime import datetime, timedelta
import gradio as gr
import matplotlib.pyplot as plt
import pandas as pd
import seaborn as sns
import duckdb
import logging
from tabs.tokens_votes_dist import (
    get_based_tokens_distribution,
    get_based_votes_distribution,
)
from tabs.dist_gap import (
    get_distribution_plot,
    get_correlation_map,
    get_kde_with_trades,
)


def get_logger():
    logger = logging.getLogger(__name__)
    logger.setLevel(logging.DEBUG)
    # stream handler and formatter
    stream_handler = logging.StreamHandler()
    stream_handler.setLevel(logging.DEBUG)
    formatter = logging.Formatter(
        "%(asctime)s - %(name)s - %(levelname)s - %(message)s"
    )
    stream_handler.setFormatter(formatter)
    logger.addHandler(stream_handler)
    return logger


logger = get_logger()


def prepare_data():
    """
    Get all data from the parquet files
    """
    logger.info("Getting all data")
    con = duckdb.connect(":memory:")

    # Query to fetch invalid trades data
    query = f"""
    SELECT *
    FROM read_parquet('./live_data/markets_live_data.parquet')
    """
    df = con.execute(query).fetchdf()
    df["sample_datetime"] = df["sample_timestamp"].apply(
        lambda x: datetime.fromtimestamp(x)
    )
    df["opening_datetime"] = df["openingTimestamp"].apply(
        lambda x: datetime.fromtimestamp(int(x))
    )
    df["days_to_resolution"] = (df["opening_datetime"] - df["sample_datetime"]).dt.days
    return df


def get_extreme_cases(live_fpmms: pd.DataFrame):
    """Function to return the id of the best and worst case according to the dist gap metric"""
    # select markets with more than 1 sample
    samples_per_market = (
        live_fpmms[["id", "sample_timestamp"]].groupby("id").count().reset_index()
    )
    markets_with_multiple_samples = list(
        samples_per_market.loc[samples_per_market["sample_timestamp"] > 1, "id"].values
    )
    selected_markets = live_fpmms.loc[
        live_fpmms["id"].isin(markets_with_multiple_samples)
    ]
    selected_markets.sort_values(by="dist_gap_perc", ascending=False, inplace=True)
    return selected_markets.iloc[-1].id, selected_markets.iloc[0].id


demo = gr.Blocks()
markets_data = prepare_data()

with demo:
    gr.HTML("<h1>Olas Predict Live Markets </h1>")
    gr.Markdown("This app shows the distributions of predictions on the live markets.")
    best_market_id, worst_market_id = get_extreme_cases(markets_data)
    with gr.Tabs():
        with gr.TabItem("💹 Probability distributions of live markets"):
            with gr.Row():
                gr.Markdown("Best case: a market with a low gap between distributions")
            with gr.Row():
                gr.Markdown(f"Market id = {best_market_id}")
            with gr.Row():
                with gr.Column(min_width=350):
                    gr.Markdown("# Evolution of outcomes probability based on tokens")
                    best_market_tokens_dist = get_based_tokens_distribution(
                        best_market_id, markets_data
                    )
                with gr.Column(min_width=350):
                    gr.Markdown("# Evolution of outcomes probability based on votes")
                    best_market_votes_dist = get_based_votes_distribution(
                        best_market_id, markets_data
                    )

            with gr.Row():
                gr.Markdown("Worst case: a market with a high distribution gap metric")
            with gr.Row():
                gr.Markdown(f"Market id = {worst_market_id}")

            with gr.Row():
                with gr.Column(min_width=350):
                    # gr.Markdown("# Evolution of outcomes probability based on tokens")
                    worst_market_tokens_dist = get_based_tokens_distribution(
                        worst_market_id, markets_data
                    )
                with gr.Column(min_width=350):
                    worst_market_votes_dist = get_based_votes_distribution(
                        worst_market_id, markets_data
                    )

        with gr.TabItem("📏 Distribution gap metric"):
            with gr.Row():
                gr.Markdown(
                    "This metric measures the difference between the probability distribution based on the tokens distribution and the one based on the votes distribution"
                )
            with gr.Row():
                gr.Markdown("# Density distribution")
            with gr.Row():
                kde_plot = get_distribution_plot(markets_data)

            with gr.Row():
                gr.Markdown("# Relationship with number of trades")

            with gr.Row():
                kde_trades_plot = get_kde_with_trades(markets_data)

            with gr.Row():
                gr.Markdown("# Correlation analysis between variables")

            with gr.Row():
                correlation_plot = get_correlation_map(markets_data)

    demo.queue(default_concurrency_limit=40).launch()