Get Binance data / store into a TSDB ? / get data as dataframe

[c0indev3l]

Hello,

for storing historical data you may be interested in using a timeseries database.

Here is some code to download data from Binance

import datetime
from binance_historical_data import BinanceDataDumper

data_dumper = BinanceDataDumper(
    path_dir_where_to_dump=".",
    asset_class="spot",  # spot, um, cm
    data_type="klines",  # aggTrades, klines, trades
    data_frequency="1h",
)
data_dumper.dump_data(
    tickers="BTCUSDT",
    date_start=datetime.date(2023, 7, 1),
    date_end=None,
    is_to_update_existing=False,
)

using https://pypi.org/project/binance-historical-data/

Store data into an InfluxDB database

from influxdb_client import InfluxDBClient, WriteOptions
import datetime
import pandas as pd

data_source = "binance"
asset_class = "spot"  # spot, um, cm
storage_frequency = "daily"  # daily, monthly
data_type = "klines"  # aggTrades, klines, trades
symbol = "BTCUSDT"
data_frequency = "1h"
dt = datetime.date(2023, 7, 1)
fname = f"C:\\Users\\w4c\\data\\{data_source}\\{asset_class}\\{storage_frequency}\\{data_type}\\{symbol}\\{data_frequency}\\{symbol}-{data_frequency}-{dt.year}-{dt.month:02}-{dt.day:02}.csv"
with InfluxDBClient.from_env_properties() as client:
    columns = [
        "OpenTime",
        "Open",
        "High",
        "Low",
        "Close",
        "Volume",
        "CloseTime",
        "Quote asset volume",
        "Number of trades",
        "Taker buy base asset volume",
        "Taker buy quote asset volume",
        "Ignore",
    ]
    for df in pd.read_csv(fname, chunksize=1_000, names=columns):
        # for col in ["OpenTime", "CloseTime"]:
        for col in ["OpenTime"]:
            df[col] = pd.to_datetime(df[col], unit="ms")
        df["CloseTime"] *= 1_000_000.0
        df["data_source"] = data_source
        df["asset_class"] = asset_class
        df["data_type"] = data_type
        df["data_frequency"] = data_frequency
        df["symbol"] = symbol
        print(df)
        print(df.dtypes)
        with client.write_api() as write_api:
            try:
                write_api.write(
                    record=df,
                    bucket="data",
                    data_frame_measurement_name="crypto",
                    data_frame_tag_columns=[
                        "data_source",
                        "asset_class",
                        "data_type",
                        "symbol",
                        "data_frequency",
                    ],
                    data_frame_timestamp_column="OpenTime",
                )
            except Exception as e:
                print(e)

using https://github.com/influxdata/influxdb-client-python

Retrieve data as Pandas DataFrame

from influxdb_client import InfluxDBClient, WriteOptions
import pandas as pd

pd.options.display.max_rows = 10
pd.options.display.max_columns = 20

data_source = "binance"
asset_class = "spot"  # spot, um, cm
data_type = "klines"  # aggTrades, klines, trades
symbol = "BTCUSDT"
data_frequency = "1h"

dt_from = pd.to_datetime("2023-07-01")
dt_to = pd.to_datetime("2023-07-02")

ts_from = int(dt_from.timestamp())
ts_to = int(dt_to.timestamp())

# query = 'from(bucket:\"data\") |> range(start:-30d)'
query = f"""from(bucket:"data") 
|> range(start: {ts_from}, stop: {ts_to}) 
|> filter(fn: (r) => r.data_source == "{data_source}"
   and r.asset_class == "{asset_class}" 
   and r.data_type == "{data_type}" 
   and r.symbol == "{symbol}" 
   and r.data_frequency == "{data_frequency}"
)
|> pivot(rowKey:["_time"], columnKey: ["_field"], valueColumn: "_value")"""

with InfluxDBClient.from_env_properties() as client:
    df = client.query_api().query_data_frame(query=query)
    if len(df) > 0:
        df.drop(columns=["result", "table", "_start", "_stop"], inplace=True)
        df.rename(columns={"_time": "OpenTime"}, inplace=True)
        df["CloseTime"] = pd.to_datetime(df["CloseTime"])
    print(df)

I'm still facing an issue https://github.com/influxdata/influxdb-client-python/issues/592

Maybe an other TSDB should be considered ? TimescaleDB for example.

An other approach could be to simply store data as Parquet or Feather files (or an other format) into a hierarchical directory

Kind regards

[ffavero]

nice, right now it's plain tsv files, I was thinking to index the files to be able to use cached data if overlapping with the requested time window, but I'd gladly uses the correct tools if already existing

[c0indev3l]

data pipelines / ETL (extract transform and load) is probably the way to follow. Unfortunately I'm not DevOps. Docker / docker-compose is probably also required (and I'm quite beginer in this part also)

[ffavero]

but the data source is working already, it's not optimized, but it useable to some extents, there are many other aspect of the tools than needs to be improved/fixed. Are you running/testing the tool?

[c0indev3l]

I'm considering it. But haven't used it currently