ENH: Extract strategy performance method compute_stats

[crazy25000]

Resolves #194
Resolves #71

Extracted methods compute_drawdown_duration_peaks, compute_stats, and class _Stats from backtesting.py to a new file so that they're available to users.

This will give us the ability to explicitly get performance metrics and have direct access to the data.

TODO:

• Update unit tests
• flake8 passes
• mypy passes
• Add long/short trade statistics
  • Need info about which metrics we want (pips, duration, wins, losses...etc)
• Update documentation

[kernc]

• Add long/short trade statistics
  • Need info about which metrics we want (pips, duration, wins, losses...etc)

I'm not in favor of supplying extra new keys in the stats series. It's long as it is.

Instead the idea is, the curious user will be able to use the newly exposed backtesting.lib.compute_stats() to compute stats on any subset of trades they are interested in:

stats = bt.run()

long_trades = stats._trades.query('Size > 0')
long_stats = compute_stats(long_trades, stats._equity_curve.Equity, data)

[kernc]

long_stats = compute_stats(long_trades, stats._equity_curve.Equity, data)

Maybe even better to expose backtesting.lib.compute_stats() such that it takes simply

compute_stats(trades: pd.DataFrame,  # trades subset
              stats: pd.Series,      # previous stats
              risk_free_rate: float = 0) -> pd.Series

and obtain equity and OHLC data simply from stats passed in.

[kernc]

obtain equity and OHLC data simply from stats passed in

Particularly OHLC data is potentially too heavy (e.g. pickling for multiprocessing) to bake into returned stats. Better pass that separately.

[crazy25000]

• Add long/short trade statistics

  • Need info about which metrics we want (pips, duration, wins, losses...etc)

I'm not in favor of supplying extra new keys in the stats series. It's long as it is.

Instead the idea is, the curious user will be able to use the newly exposed backtesting.lib.compute_stats() to compute stats on any subset of trades they are interested in:

stats = bt.run()

long_trades = stats._trades.query('Size > 0')
long_stats = compute_stats(long_trades, stats._equity_curve.Equity, data)

I agree, it would be better to provide the information and let users calculate metrics they need. I wasn't sure if we had to add the long/short stats based on the original issue conversation.

[kernc]

Good. Now let's expose+document this function publicly in backtesting.lib. The user will be able to pass trades: pd.DataFrame as obtained by this exact method (stats._trades), so you have to figure out a way for that.

[crazy25000]

@kernc passing stats._strategy._broker.closed_trades makes it compatible as is. If we want to pass stats._trades, we'd need to convert the DataFrame to a list of Trade.

[crazy25000]

Oops 😛

[kernc]

@crazy25000 Apologies for the late review. 🌞 Many thanks for your work on this! 🥮