mirror of
https://github.com/freqtrade/freqtrade.git
synced 2024-11-14 20:23:57 +00:00
89 lines
3.1 KiB
Python
89 lines
3.1 KiB
Python
import logging
|
|
from pathlib import Path
|
|
|
|
import pandas as pd
|
|
|
|
from freqtrade.constants import DATETIME_PRINT_FORMAT, DEFAULT_TRADES_COLUMNS, Config
|
|
from freqtrade.data.converter.trade_converter import (
|
|
trades_convert_types,
|
|
trades_df_remove_duplicates,
|
|
)
|
|
from freqtrade.data.history import get_datahandler
|
|
from freqtrade.enums import TradingMode
|
|
from freqtrade.exceptions import OperationalException
|
|
from freqtrade.plugins.pairlist.pairlist_helpers import expand_pairlist
|
|
from freqtrade.resolvers import ExchangeResolver
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
KRAKEN_CSV_TRADE_COLUMNS = ["timestamp", "price", "amount"]
|
|
|
|
|
|
def import_kraken_trades_from_csv(config: Config, convert_to: str):
|
|
"""
|
|
Import kraken trades from csv
|
|
"""
|
|
if config["exchange"]["name"] != "kraken":
|
|
raise OperationalException("This function is only for the kraken exchange.")
|
|
|
|
datadir: Path = config["datadir"]
|
|
data_handler = get_datahandler(datadir, data_format=convert_to)
|
|
|
|
tradesdir: Path = config["datadir"] / "trades_csv"
|
|
exchange = ExchangeResolver.load_exchange(config, validate=False)
|
|
# iterate through directories in this directory
|
|
data_symbols = {p.stem for p in tradesdir.rglob("*.csv")}
|
|
|
|
# create pair/filename mapping
|
|
markets = {
|
|
(m["symbol"], m["altname"])
|
|
for m in exchange.markets.values()
|
|
if m.get("altname") in data_symbols
|
|
}
|
|
logger.info(f"Found csv files for {', '.join(data_symbols)}.")
|
|
|
|
if pairs_raw := config.get("pairs"):
|
|
pairs = expand_pairlist(pairs_raw, [m[0] for m in markets])
|
|
markets = {m for m in markets if m[0] in pairs}
|
|
if not markets:
|
|
logger.info(f"No data found for pairs {', '.join(pairs_raw)}.")
|
|
return
|
|
logger.info(f"Converting pairs: {', '.join(m[0] for m in markets)}.")
|
|
|
|
for pair, name in markets:
|
|
logger.debug(f"Converting pair {pair}, files */{name}.csv")
|
|
dfs = []
|
|
# Load and combine all csv files for this pair
|
|
for f in tradesdir.rglob(f"{name}.csv"):
|
|
df = pd.read_csv(f, names=KRAKEN_CSV_TRADE_COLUMNS)
|
|
if not df.empty:
|
|
dfs.append(df)
|
|
|
|
# Load existing trades data
|
|
if not dfs:
|
|
# edgecase, can only happen if the file was deleted between the above glob and here
|
|
logger.info(f"No data found for pair {pair}")
|
|
continue
|
|
|
|
trades = pd.concat(dfs, ignore_index=True)
|
|
del dfs
|
|
|
|
trades.loc[:, "timestamp"] = trades["timestamp"] * 1e3
|
|
trades.loc[:, "cost"] = trades["price"] * trades["amount"]
|
|
for col in DEFAULT_TRADES_COLUMNS:
|
|
if col not in trades.columns:
|
|
trades.loc[:, col] = ""
|
|
trades = trades[DEFAULT_TRADES_COLUMNS]
|
|
trades = trades_convert_types(trades)
|
|
|
|
trades_df = trades_df_remove_duplicates(trades)
|
|
del trades
|
|
logger.info(
|
|
f"{pair}: {len(trades_df)} trades, from "
|
|
f"{trades_df['date'].min():{DATETIME_PRINT_FORMAT}} to "
|
|
f"{trades_df['date'].max():{DATETIME_PRINT_FORMAT}}"
|
|
)
|
|
|
|
data_handler.trades_store(pair, trades_df, TradingMode.SPOT)
|