Data Feeds

backtrader comes with a set of Data Feed parsers (at the time of writing all CSV Based) to let you load data from different sources.

  • Yahoo (online or already saved to a file)
  • VisualChart (see www.visualchart.com
  • Backtrader CSV (own cooked format for testing)
  • Generic CSV support

From the Quickstart guide it should be clear that you add data feeds to a Cerebro instance. The data feeds will later be available to the different strategies in:

  • An array self.datas (insertion order)
  • Alias to the array objects:
    • self.data and self.data0 point to the first element
    • self.dataX points to elements with index X in the array

A quick reminder as to how the insertion works:

import backtrader as bt
import backtrader.feeds as btfeeds

data = btfeeds.YahooFinanceCSVData(dataname='wheremydatacsvis.csv')

cerebro = bt.Cerebro()

cerebro.adddata(data)  # a 'name' parameter can be passed for plotting purposes

Data Feeds Common parameters

This data feed can download data directly from Yahoo and feed into the system.

Parameters:

  • dataname (default: None) MUST BE PROVIDED

    The meaning varies with the data feed type (file location, ticker, ...)

  • name (default: ‘’)

    Meant for decorative purposes in plotting. If not specified it may be derived from dataname (example: last part of a file path)

  • fromdate (default: mindate)

    Python datetime object indicating that any datetime prior to this should be ignored

  • todate (default: maxdate)

    Python datetime object indicating that any datetime posterior to this should be ignored

  • timeframe (default: TimeFrame.Days)

    Potential values: Ticks, Seconds, Minutes, Days, Weeks, Months and Years

  • compression (default: 1)

    Number of actual bars per bar. Informative. Only effective in Data Resampling/Replaying.

  • sessionstart (default: None)

    Indication of session starting time for the data. May be used by classes for purposes like resampling

  • sessionend (default: None)

    Indication of session ending time for the data. May be used by classes for purposes like resampling

CSV Data Feeds Common parameters

Parameters (additional to the common ones):

  • headers (default: True)

    Indicates if the passed data has an initial headers row

  • separator (default: ”,”)

    Separator to take into account to tokenize each of the CSV rows

GenericCSVData

This class exposes a generic interface allowing parsing mostly every CSV file format out there.

Parses a CSV file according to the order and field presence defined by the parameters

Specific parameters (or specific meaning):

  • dataname

    The filename to parse or a file-like object

  • datetime (default: 0) column containing the date (or datetime) field

  • time (default: -1) column containing the time field if separate from the datetime field (-1 indicates it’s not present)

  • open (default: 1) , high (default: 2), low (default: 3), close (default: 4), volume (default: 5), openinterest (default: 6)

    Index of the columns containing the corresponding fields

    If a negative value is passed (example: -1) it indicates the field is not present in the CSV data

  • nullvalue (default: float(‘NaN’))

    Value that will be used if a value which should be there is missing (the CSV field is empty)

  • dtformat (default: %Y-%m-%d %H:%M:%S)

    Format used to parse the datetime CSV field

  • tmformat (default: %H:%M:%S)

    Format used to parse the time CSV field if “present” (the default for the “time” CSV field is not to be present)

An example usage covering the following requirements:

  • Limit input to year 2000
  • HLOC order rather than OHLC
  • Missing values to be replaced with zero (0.0)
  • Daily bars are provided and datetime is just the day with format YYYY-MM-DD
  • No openinterest column is present

The code:

import datetime
import backtrader as bt
import backtrader.feeds as btfeeds

...
...

data = btfeeds.GenericCSVData(
    dataname='mydata.csv',

    fromdate=datetime.datetime(2000, 1, 1),
    todate=datetime.datetime(2000, 12, 31),

    nullvalue=0.0,

    dtformat=('%Y-%m-%d'),

    datetime=0,
    high=1,
    low=2,
    open=3,
    close=4,
    volume=5,
    openinterest=-1
)

...

Slightly modified requirements:

  • Limit input to year 2000
  • HLOC order rather than OHLC
  • Missing values to be replaced with zero (0.0)
  • Intraday bars are provided, with separate date and time columns - Date has format YYYY-MM-DD - Time has format HH.MM.SS (instead of the usual HH:MM:SS)
  • No openinterest column is present

The code:

import datetime
import backtrader as bt
import backtrader.feeds as btfeed

...
...

data = btfeeds.GenericCSVData(
    dataname='mydata.csv',

    fromdate=datetime.datetime(2000, 1, 1),
    todate=datetime.datetime(2000, 12, 31),

    nullvalue=0.0,

    dtformat=('%Y-%m-%d'),
    tmformat=('%H.%M.%S'),

    datetime=0,
    time=1,
    high=2,
    low=3,
    open=4,
    close=5,
    volume=6,
    openinterest=-1
)

This can also be made permanent with subclassing:

import datetime
import backtrader.feeds as btfeed

class MyHLOC(btfreeds.GenericCSVData):

  params = (
    ('fromdate', datetime.datetime(2000, 1, 1)),
    ('todate', datetime.datetime(2000, 12, 31)),
    ('nullvalue', 0.0),
    ('dtformat', ('%Y-%m-%d')),
    ('tmformat', ('%H.%M.%S')),

    ('datetime', 0),
    ('time', 1),
    ('high', 2),
    ('low', 3),
    ('open', 4),
    ('close', 5),
    ('volume', 6),
    ('openinterest', -1)
)

This new class can be reused now by just providing the dataname:

data = btfeeds.MyHLOC(dataname='mydata.csv')