scripts/utils.py

import pandas as pd
from pathlib import Path
from datetime import datetime, timedelta
import gzip
import shutil
import os
from huggingface_hub import hf_hub_download

SCRIPTS_DIR = Path(__file__).parent
ROOT_DIR = SCRIPTS_DIR.parent
TMP_DIR = ROOT_DIR / "tmp"


def get_traders_family(row: pd.DataFrame) -> str:
    if row.staking == "non_agent":
        return "non_agent"
    elif row.market_creator == "pearl":
        return "pearl_agent"
    # quickstart
    return "quickstart_agent"


def get_current_week():
    current_date = datetime.now()

    # Get the start and end dates of the current week (starting on Sunday)
    current_week_start = current_date - timedelta(days=current_date.weekday() + 1)
    return current_week_start.strftime("%b-%d-%Y")


def get_next_week():
    current_date = datetime.now()
    next_week_start = current_date + timedelta(days=7 - (current_date.weekday() + 1))
    return next_week_start.strftime("%b-%d-%Y")


def load_all_data():

    # all trades profitability
    # Download the compressed file
    gz_filepath_trades = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="all_trades_profitability.parquet.gz",
        repo_type="dataset",
    )

    parquet_filepath_trades = gz_filepath_trades.replace(".gz", "")
    parquet_filepath_trades = parquet_filepath_trades.replace("all", "")

    with gzip.open(gz_filepath_trades, "rb") as f_in:
        with open(parquet_filepath_trades, "wb") as f_out:
            shutil.copyfileobj(f_in, f_out)

    # Now read the decompressed parquet file
    df1 = pd.read_parquet(parquet_filepath_trades)

    # closed_markets_div
    closed_markets_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="closed_markets_div.parquet",
        repo_type="dataset",
    )
    df2 = pd.read_parquet(closed_markets_df)

    # daily_info
    daily_info_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="daily_info.parquet",
        repo_type="dataset",
    )
    df3 = pd.read_parquet(daily_info_df)

    # unknown traders
    unknown_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="unknown_traders.parquet",
        repo_type="dataset",
    )
    df4 = pd.read_parquet(unknown_df)

    # retention activity
    gz_file_path_ret = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="retention_activity.parquet.gz",
        repo_type="dataset",
    )
    parquet_file_path_ret = gz_file_path_ret.replace(".gz", "")

    with gzip.open(gz_file_path_ret, "rb") as f_in:
        with open(parquet_file_path_ret, "wb") as f_out:
            shutil.copyfileobj(f_in, f_out)
    df5 = pd.read_parquet(parquet_file_path_ret)
    # os.remove(parquet_file_path_ret)

    # active_traders.parquet
    active_traders_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="active_traders.parquet",
        repo_type="dataset",
    )
    df6 = pd.read_parquet(active_traders_df)

    # weekly_mech_calls.parquet
    all_mech_calls_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="weekly_mech_calls.parquet",
        repo_type="dataset",
    )
    df7 = pd.read_parquet(all_mech_calls_df)

    # daa for quickstart and pearl
    daa_qs_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="latest_result_DAA_QS.parquet",
        repo_type="dataset",
    )
    df8 = pd.read_parquet(daa_qs_df)

    daa_pearl_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="latest_result_DAA_Pearl.parquet",
        repo_type="dataset",
    )
    df9 = pd.read_parquet(daa_pearl_df)
    # Read weekly_avg_roi_pearl_agents.parquet
    weekly_avg_roi_pearl_agents = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="weekly_avg_roi_pearl_agents.parquet",
        repo_type="dataset",
    )
    df10 = pd.read_parquet(weekly_avg_roi_pearl_agents)

    # two_weeks_avg_roi_pearl_agents.parquet
    two_weeks_avg_roi_pearl_agents = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="two_weeks_avg_roi_pearl_agents.parquet",
        repo_type="dataset",
    )
    df11 = pd.read_parquet(two_weeks_avg_roi_pearl_agents)

    # read traders_weekly_metrics.parquet file
    traders_weekly_metrics_df = hf_hub_download(
        repo_id="valory/Olas-predict-dataset",
        filename="traders_weekly_metrics.parquet",
        repo_type="dataset",
    )
    df12 = pd.read_parquet(traders_weekly_metrics_df)
    return df1, df2, df3, df4, df5, df6, df7, df8, df9, df10, df11, df12


def prepare_data():

    (
        all_trades,
        closed_markets,
        daily_info,
        unknown_traders,
        retention_df,
        active_traders,
        all_mech_calls,
        daa_qs_df,
        daa_pearl_df,
        weekly_avg_roi_pearl_agents,
        two_weeks_avg_roi_pearl_agents,
        traders_weekly_metrics_df,
    ) = load_all_data()
    all_trades["creation_timestamp"] = all_trades["creation_timestamp"].dt.tz_convert(
        "UTC"
    )
    all_trades = all_trades.sort_values(by="creation_timestamp", ascending=True)
    all_trades["creation_date"] = all_trades["creation_timestamp"].dt.date

    # nr-trades variable
    volume_trades_per_trader_and_market = (
        all_trades.groupby(["trader_address", "title"])["roi"]
        .count()
        .reset_index(name="nr_trades_per_market")
    )

    traders_data = pd.merge(
        all_trades, volume_trades_per_trader_and_market, on=["trader_address", "title"]
    )
    daily_info["creation_date"] = daily_info["creation_timestamp"].dt.date
    unknown_traders["creation_date"] = unknown_traders["creation_timestamp"].dt.date
    active_traders["creation_date"] = active_traders["creation_timestamp"].dt.date
    # adding the trader family column
    traders_data["trader_family"] = traders_data.apply(
        lambda x: get_traders_family(x), axis=1
    )
    # print(traders_data.head())

    traders_data = traders_data.sort_values(by="creation_timestamp", ascending=True)
    unknown_traders = unknown_traders.sort_values(
        by="creation_timestamp", ascending=True
    )
    traders_data["month_year_week"] = (
        traders_data["creation_timestamp"]
        .dt.to_period("W")
        .dt.start_time.dt.strftime("%b-%d-%Y")
    )
    unknown_traders["month_year_week"] = (
        unknown_traders["creation_timestamp"]
        .dt.to_period("W")
        .dt.start_time.dt.strftime("%b-%d-%Y")
    )
    closed_markets["month_year_week"] = (
        closed_markets["opening_datetime"]
        .dt.to_period("W")
        .dt.start_time.dt.strftime("%b-%d-%Y")
    )

    # prepare the daa dataframes
    daa_pearl_df["day"] = pd.to_datetime(
        daa_pearl_df["day"], format="%Y-%m-%d 00:00:00.000 UTC"
    )
    daa_qs_df["day"] = pd.to_datetime(
        daa_qs_df["day"], format="%Y-%m-%d 00:00:00.000 UTC"
    )
    daa_pearl_df["day"] = daa_pearl_df["day"].dt.tz_localize("UTC")
    daa_qs_df["day"] = daa_qs_df["day"].dt.tz_localize("UTC")
    daa_qs_df["tx_date"] = pd.to_datetime(daa_qs_df["day"]).dt.date
    daa_pearl_df["tx_date"] = pd.to_datetime(daa_pearl_df["day"]).dt.date
    daa_pearl_df["seven_day_trailing_avg"] = pd.to_numeric(
        daa_pearl_df["seven_day_trailing_avg"], errors="coerce"
    )
    daa_pearl_df["seven_day_trailing_avg"] = daa_pearl_df[
        "seven_day_trailing_avg"
    ].round(2)
    daa_qs_df["seven_day_trailing_avg"] = pd.to_numeric(
        daa_qs_df["seven_day_trailing_avg"], errors="coerce"
    )
    daa_qs_df["seven_day_trailing_avg"] = daa_qs_df["seven_day_trailing_avg"].round(2)
    return (
        traders_data,
        closed_markets,
        daily_info,
        unknown_traders,
        retention_df,
        active_traders,
        all_mech_calls,
        daa_qs_df,
        daa_pearl_df,
        weekly_avg_roi_pearl_agents,
        two_weeks_avg_roi_pearl_agents,
        traders_weekly_metrics_df,
    )