General Code/Help

• R

  PandasData won't read a datetime column
  • randomshinichi

  2
  2
  Posts
  31
  Views

  B

  Even if not obvious, a 2 liner of the original data (you can fake it if you wish) would be key to actually try to reproduce your behavior.

  All those ipdb statements are only confusing, giving information about many things which have nothing to do with the potential problem.

  In any case this is probably due to a recent pull request was issued to remove the usage of ix (deprecated in the latest versions of Pandas), which is replaced with either loc (label based) or iloc (numeric based) and which didn't catch all use cases (In most occasions the datetime timestamps are the index of the dataframe)

  The latest development commit contains some code to correct the situation and simplify the autodetection:

  https://github.com/mementum/backtrader/commit/753012adf3590d0d43de40565d16bd7f61fa8861
• P

  quandl example
  • pookeye

  6
  6
  Posts
  109
  Views

  B

  See Docs - Data Feeds Reference and the reference for the Quandl data feed.

  It could be (unknown here) that what you have before the / is the dataset and you have to pass it using that parameter, dataset=CURRFX

• M

  Oanda v20 API
  • Matthias Janssen

  1
  1
  Posts
  31
  Views

  No one has replied

• P

  a bug in code?
  • phoenix.che

  2
  2
  Posts
  42
  Views

  B

  The code is for sure prone to contain bugs, but rather than isolated code lines

  Do you have a small test sample that shows that there is a bug there?
• B

  IB Live Data Feed: differences between resampled data and backfilled data
  • borodiliz

  4
  4
  Posts
  66
  Views

  B

  @borodiliz said in IB Live Data Feed: differences between resampled data and backfilled data:

  My conclusion: First rtbar is not taken into account

  May of course be. Will be looked into. Thank you for the analysis.

• 

  Getting the current date in a strategy's next() method
  • Curtis Miller

  4
  4
  Posts
  61
  Views

  B

  There is a light python2-3 adaptation layer inside backtrader in backtrader.utils.py3, mostly to avoid importing six or similar packages. zip is including as a generator for Python2 (itertools.izip) to match the Python3 style.

  Inside the core, the adaptation layer is used and if a generator is not the needed result it will be simply wrapped in a list(generator). This makes things consistent and allows the package to work with Python2/3

• R

  How to get an array containing the current and past values of an indicator.
  • Rodrigo Pinto

  4
  4
  Posts
  56
  Views

  B

  An explanation has been added to:

  Community - FAQ Docs - Platform Concepts
• C

  Indicator Period Days or Trading Days?
  • cwse

  4
  4
  Posts
  56
  Views

  C

  Cool thanks guys

• C

  Save analyzer Output to Log
  • cwse

  4
  4
  Posts
  54
  Views

  B

  The actual output produced by an analyzer is free. It is regularly a dict but must not be. It is whatever is returned by get_analysis and as such free format.

  The Writer, which as pointed out by @ab_trader writes everything to a file, tries to take into account as much as possible to produce a sensible output, but this may not work with all possible return values of get_analysis

• B

  PSAR indicator and multiple timeframes
  • borodiliz

  21
  21
  Posts
  463
  Views

  B

  Thanks @backtrader. That will cover most cases. Maybe the SAREXT indicator should also be added?

  In my case I'm working with a very large datasheet so to avoid performance issues I've coded something like:

  def next(self): self.psar0._lookback = min(100, len(self.data0))

  so I have a "dynamic" _lookback from 1 to 100

  And why 100?... newbie here! just by observation, it seems to be a good value to calculate a real ParabolicSAR value.

• S

  [Cerebro] Unresolved reference 'msg'
  • Sai Jamic

  5
  5
  Posts
  76
  Views

  B

  It's already in the latest release Community - Release 1.9.51.121

• 

  Trades don't happen until WAY late (NO, THEY DON'T)
  • Curtis Miller

  6
  6
  Posts
  119
  Views

  B

  That's exactly the implemented behavior: no-op and that's why the strategy remains in the prenext phase, moving only to the next (yes-op) when data is available.

  You may forward each prenext call to next and always be in the yes-op mode.

• C

  Broker.getvalue() = nan?
  • cwse

  5
  5
  Posts
  89
  Views

  C

  Alright thanks. I filled all NaNs with '0' and issue seems to have resolved

• 

  Plotting JUST the account's value
  • Curtis Miller

  4
  4
  Posts
  91
  Views

  B

  The actual documented way to do it:

  cerebro.addobservermulti(bt.observers.BuySell)

  The source code uses internal methods to do things which may break at any time.

  See Docs - Cerebro and look for addoservermulti (the reference is close to the end) It will add the observer as many times as data feeds are present in the system (with each observer in charge of the things happening on one data feed)

• H

  Json data and general programming question
  • hippylover

  5
  5
  Posts
  94
  Views

  B

  You can create your own data feed and fill the values from your JSON data. See

  Docs - Binary Datafeed Development

  You basically subclass and override _load to deliver your set of OHLC prices each time.

  @hippylover said in Json data and general programming question:

  Also, does backtester only take open and close data or can i feed it just last bid for example?
  @Curtis-Miller said in Json data and general programming question:
  I'm also curious about data that isn't OHLC (like tick data).

  You can extend the data feed to pass any values you wish. See: Docs - Extending a Datafeed

  In any case, bid-ask are values tied to ticks and not to larger timeframes (mostly resampled) which would only have the latest of the bid-ask values to act upon or else the next incoming. This platform doesn't use bid-ask for anything.

  @hippylover said in Json data and general programming question:

  Also, if i want each "period" for calculating/plotting ema/sma/etc to be 30-minutes, is it possible to set this or do i have to feed it half-hour data?

  With no time sample it is impossible to know what timeframe your data is in. But you can resample the data (and even replay it). See Docs - Data Resampling

  You may also want to read the FAQ, to remind yourself of setting the right timeframe and compression for your data. See Community - FAQ

• B

  [BC Break] IndexError: index out of range after commit cdb80f8472fc1aeb05d84c5b52b86412e3f89632
  • borodiliz

  4
  4
  Posts
  61
  Views

  B

  You can always post anything you may consider relevant, but in general a development branch should only be used to check the functionality of a specific commit which may solve something of your interest.

• D

  [Benchmark Observer] Issue when benchmarking with resampled data
  • dimitar-petrov

  7
  7
  Posts
  128
  Views

  D

  Works for me,
  Thanks

• D

  KeyError when using multiple dataframes
  • derdav3

  3
  3
  Posts
  76
  Views

  B

  The answer was known because this was already a detected use case and was in the works. You may alternatively try the latest development version which allows plotmaster to reference to the data it belongs to (which allows for a clean loop like the one you were trying)

  https://github.com/mementum/backtrader/commit/dd5c0c87b1895a7428b35930cffe8d364d21a271
• K

  Cancel Order
  • KT

  9
  9
  Posts
  131
  Views

  K

  Thanks @backtrader !

• T

  My (different) data didn't start/end at the same date.
  • tempurak

  4
  4
  Posts
  92
  Views

  B

  This is by design. Each object has a minimum period and while that minimum period is not met, the objects dependent on those cannot go into next and remain in next. For data feeds the minimum period is simply1`, i.e.: they have started delivering data.

  Being the reason as simple as:

  def next(self): ... print(self.data0.close[0]) # ok ... starts earlier as self.data1 and has delivered print(self.data1.close[0]) # NOK ... kaboom ... exception raised if data1 hadn't started to deliver but it has ... ...

  The idea is that code NEVER breaks inside next. There is no code guard around self.data1 checking if it has already delivered, because it has. It is guaranteed in next that all objects already deliver.

  See:

  Docs - Operating the platform Docs - Strategy

  You can always forward each call to prenext to next, but you will have to manually check the length of data1 (and associated indicators) to make sure the code doesn't break.

  def next(self): ... print(self.data0.close[0]) # ok ... starts earlier as self.data1 and has delivered if len(self.data1): print(self.data1.close[0]) # guard placed because it may not have delivered ... ...