"""Analyze the trade performance of algorithm.
Calculate success/fail rate of trades and plot success distribution.
Example analysis include:
- Table: Summary of all trades
- Graph: Trade won/lost distribution
- Timeline: Analysis of each individual trades made
.. note ::
A lot of this code has been lifted off from trading-strategy package
where it had to deal with different trading frameworks.
It could be simplified greatly now.
"""
import datetime
import warnings
import enum
import logging
from dataclasses import dataclass, field
from typing import List, Dict, Iterable, Optional, Tuple, Callable, Set
import numpy as np
import pandas as pd
from IPython.core.display_functions import display
from IPython.display import HTML
from dataclasses_json import dataclass_json, config
from statistics import median
from tradeexecutor.state.position import TradingPosition
from tradeexecutor.state.portfolio import Portfolio
from tradeexecutor.state.trade import TradeExecution, TradeType
from tradeexecutor.state.types import USDollarPrice, Percent
from tradeexecutor.utils.format import calculate_percentage
from tradeexecutor.utils.timestamp import json_encode_timedelta, json_decode_timedelta
from tradingstrategy.timebucket import TimeBucket
from tradingstrategy.exchange import Exchange
from tradingstrategy.pair import PandasPairUniverse
from tradingstrategy.types import PrimaryKey, USDollarAmount
from tradingstrategy.utils.format import format_value, format_price, format_duration_days_hours_mins, \
format_percent_2_decimals
from tradeexecutor.utils.summarydataframe import as_dollar, as_integer, create_summary_table, as_percent, as_duration, as_bars
try:
# DeprecationWarning: Importing display from IPython.core.display is deprecated since IPython 7.14, please import from IPython display
with warnings.catch_warnings():
import quantstats as qs
HAS_QUANTSTATS = True
except Exception:
HAS_QUANTSTATS = False
logger = logging.getLogger(__name__)
[docs]@dataclass_json
@dataclass(slots=True)
class TradeSummary:
"""Some generic statistics over all the trades
TDOO: Cleam this up
"""
won: int
lost: int
zero_loss: int
stop_losses: int
undecided: int
realised_profit: USDollarAmount
open_value: USDollarAmount
uninvested_cash: USDollarAmount
initial_cash: USDollarAmount
extra_return: USDollarAmount
duration: Optional[datetime.timedelta] = field(metadata=config(
encoder=json_encode_timedelta,
decoder=json_decode_timedelta,
))
average_winning_trade_profit_pc: Optional[float] # position
average_losing_trade_loss_pc: Optional[float] # position
biggest_winning_trade_pc: Optional[float] # position
biggest_losing_trade_pc: Optional[float] # position
average_duration_of_winning_trades: datetime.timedelta = field(metadata=config(
encoder=json_encode_timedelta,
decoder=json_decode_timedelta,
)) # position
average_duration_of_losing_trades: datetime.timedelta = field(metadata=config(
encoder=json_encode_timedelta,
decoder=json_decode_timedelta,
)) # position
time_bucket: Optional[TimeBucket] = None
# these stats calculate in post-init, so init=False
total_positions: int = field(init=False)
win_percent: float = field(init=False)
lost_percent: float = field(init=False)
return_percent: float = field(init=False)
annualised_return_percent: float = field(init=False)
all_stop_loss_percent: float = field(init=False)
# (total stop losses)/(lost positions)
# can be > 1 if there are more stop losses than lost positions
# TODO this is a confusing metric, more intuitive is (losing stop losses)/(total_stop_losses)
# and (winning stop losses)/(total_stop_losses)
lost_stop_loss_percent: float = field(init=False)
all_take_profit_percent: float = field(init=False)
# (take profits)/(won positions)
# can be > 1 if there are more take profits than won positions
# same confusion as lost_stop_loss_percent
# TODO add (winning take profits)/(total_take_profits)
# and (losing take profits)/(total_take_profits)
won_take_profit_percent: float = field(init=False)
average_net_profit: USDollarAmount = field(init=False)
end_value: USDollarAmount = field(init=False)
average_trade: Optional[float] = None # position
median_trade: Optional[float] = None # position
max_pos_cons: Optional[int] = None
max_neg_cons: Optional[int] = None
max_pullback: Optional[float] = None
max_loss_risk: Optional[float] = None
max_realised_loss: Optional[float] = None
avg_realised_risk: Optional[Percent] = None
take_profits: int = field(default=0)
trade_volume: USDollarAmount = field(default=0.0)
lp_fees_paid: Optional[USDollarPrice] = 0
lp_fees_average_pc: Optional[USDollarPrice] = 0
#: advanced users can use this property instead of the
#: provided quantstats helper methods
daily_returns: Optional[pd.Series] = None
winning_stop_losses: Optional[int] = 0
losing_stop_losses: Optional[int] = 0
winning_take_profits: Optional[int] = 0
losing_take_profits: Optional[int] = 0
winning_stop_losses_percent: Optional[float] = field(init=False)
losing_stop_losses_percent: Optional[float] = field(init=False)
winning_take_profits_percent: Optional[float] = field(init=False)
losing_take_profits_percent: Optional[float] = field(init=False)
median_win: Optional[float] = None
median_loss: Optional[float] = None
sharpe_ratio: Optional[float] = None
sortino_ratio: Optional[float] = None
profit_factor: Optional[float] = None
max_drawdown: Optional[float] = None
max_runup: Optional[float] = None
average_duration_of_zero_loss_trades: Optional[datetime.timedelta] = None
average_duration_of_all_trades: Optional[datetime.timedelta] = None
def __post_init__(self):
self.total_positions = self.won + self.lost + self.zero_loss
self.win_percent = calculate_percentage(self.won, self.total_positions)
self.lost_percent = calculate_percentage(self.lost, self.total_positions)
self.all_stop_loss_percent = calculate_percentage(self.stop_losses, self.total_positions)
self.all_take_profit_percent = calculate_percentage(self.take_profits, self.total_positions)
self.lost_stop_loss_percent = calculate_percentage(self.stop_losses, self.lost)
self.won_take_profit_percent = calculate_percentage(self.take_profits, self.won)
self.average_net_profit = calculate_percentage(self.realised_profit, self.total_positions)
self.end_value = self.open_value + self.uninvested_cash
initial_cash = self.initial_cash or 0
self.return_percent = calculate_percentage(self.end_value - initial_cash, initial_cash)
self.annualised_return_percent = calculate_percentage(self.return_percent * datetime.timedelta(days=365),
self.duration) if self.return_percent else None
self.winning_stop_losses_percent = calculate_percentage(self.winning_stop_losses, self.stop_losses)
self.losing_stop_losses_percent = calculate_percentage(self.losing_stop_losses, self.stop_losses)
self.winning_take_profits_percent = calculate_percentage(self.winning_take_profits, self.take_profits)
self.losing_take_profits_percent = calculate_percentage(self.losing_take_profits, self.take_profits)
[docs] def to_dataframe(self) -> pd.DataFrame:
"""Creates a human-readable Pandas dataframe summary table from the object."""
human_data = {
"Trading period length": as_duration(self.duration),
"Return %": as_percent(self.return_percent),
"Annualised return %": as_percent(self.annualised_return_percent),
"Cash at start": as_dollar(self.initial_cash),
"Value at end": as_dollar(self.end_value),
"Trade volume": as_dollar(self.trade_volume),
"Position win percent": as_percent(self.win_percent),
"Total positions": as_integer(self.total_positions),
"Won positions": as_integer(self.won),
"Lost positions": as_integer(self.lost),
"Stop losses triggered": as_integer(self.stop_losses),
"Stop loss % of all": as_percent(self.all_stop_loss_percent),
"Stop loss % of lost": as_percent(self.lost_stop_loss_percent),
"Winning stop losses": as_integer(self.winning_stop_losses),
"Winning stop losses percent": as_percent(self.winning_stop_losses_percent),
"Losing stop losses": as_integer(self.losing_stop_losses),
"Losing stop losses percent": as_percent(self.losing_stop_losses_percent),
"Take profits triggered": as_integer(self.take_profits),
"Take profit % of all": as_percent(self.all_take_profit_percent),
"Take profit % of won": as_percent(self.won_take_profit_percent),
"Zero profit positions": as_integer(self.zero_loss),
"Positions open at the end": as_integer(self.undecided),
"Realised profit and loss": as_dollar(self.realised_profit),
"Portfolio unrealised value": as_dollar(self.open_value),
"Extra returns on lending pool interest": as_dollar(self.extra_return),
"Cash left at the end": as_dollar(self.uninvested_cash),
"Average winning position profit %": as_percent(self.average_winning_trade_profit_pc),
"Average losing position loss %": as_percent(self.average_losing_trade_loss_pc),
"Biggest winning position %": as_percent(self.biggest_winning_trade_pc),
"Biggest losing position %": as_percent(self.biggest_losing_trade_pc),
"Average duration of winning positions": self.format_duration(self.average_duration_of_winning_trades),
"Average duration of losing positions": self.format_duration(self.average_duration_of_losing_trades),
"LP fees paid": as_dollar(self.lp_fees_paid),
"LP fees paid % of volume": as_percent(self.lp_fees_average_pc),
}
def add_prop(value, key: str, formatter: Callable):
human_data[key] = (
formatter(value)
if value is not None
else formatter(0)
)
add_prop(self.average_trade, 'Average position:', as_percent)
add_prop(self.median_trade, 'Median position:', as_percent)
add_prop(self.max_pos_cons, 'Most consecutive wins', as_integer)
add_prop(self.max_neg_cons, 'Most consecutive losses', as_integer)
add_prop(self.max_realised_loss, 'Biggest realized risk', as_percent)
add_prop(self.avg_realised_risk, 'Avg realised risk', as_percent)
add_prop(self.max_pullback, 'Max pullback of total capital', as_percent)
add_prop(self.max_loss_risk, 'Max loss risk at opening of position', as_percent)
df = create_summary_table(human_data)
return df
[docs] def display(self):
"""Create human readable summary tables and display them in IPython notebook."""
assert self.daily_returns is not None, "No daily returns data available. Remember to add state argument to calculate_summary_statistics() i.e. summary.calculate_summary_statistics(time_bucket, state). \n Otherwise you can display the old summary table with display(summary.to_dataframe())"
assert self.time_bucket is not None, "Please provide time bucket argument to calculate_summary_statistics() i.e. summary.calculate_summary_statistics(time_bucket, state). \n Otherwise you can display the old summary table with display(summary.to_dataframe())"
data1 = {
"Annualised return %": as_percent(self.annualised_return_percent),
"Lifetime return %": as_percent(self.return_percent),
"Realised PnL": as_dollar(self.realised_profit),
'Trade period': as_duration(self.duration),
}
df1 = create_summary_table(data1, "", "Returns")
data2 = {
"Total assets": as_dollar(self.end_value),
"Cash left": as_dollar(self.uninvested_cash),
"Open position value": as_dollar(self.open_value),
"Open positions": as_integer(self.undecided),
}
df2 = create_summary_table(data2, "", "Holdings")
data3 = {
'Number of positions': [
as_integer(self.won),
as_integer(self.lost),
as_integer(self.total_positions)
],
'% of total': [
as_percent(self.win_percent),
as_percent(self.lost_percent),
as_percent((self.won + self.lost)/self.total_positions)
],
'Average PnL %': [
as_percent(self.average_winning_trade_profit_pc),
as_percent(self.average_losing_trade_loss_pc),
as_percent(self.average_trade)
],
'Median PnL %': [
as_percent(self.median_win),
as_percent(self.median_loss),
as_percent(self.median_trade)],
'Biggest PnL %': [
as_percent(self.biggest_winning_trade_pc),
as_percent(self.biggest_losing_trade_pc),
as_percent(None)
],
'Average duration': [
self.format_duration(self.average_duration_of_winning_trades),
self.format_duration(self.average_duration_of_losing_trades),
self.format_duration(self.average_duration_of_all_trades)
],
'Max consecutive streak': [
as_integer(self.max_pos_cons),
as_integer(self.max_neg_cons),
as_percent(None)
],
'Max runup / drawdown': [
as_percent(self.max_runup),
as_percent(self.max_drawdown),
as_percent(None)
],
}
df3 = create_summary_table(data3, ["Winning", "Losing", "Total"], "Closed Positions")
data4 = {
'Triggered exits': [
as_integer(self.stop_losses),
as_integer(self.take_profits)
],
'Percent winning': [
as_percent(self.winning_stop_losses_percent),
as_percent(self.winning_take_profits_percent)
],
'Percent losing': [
as_percent(self.losing_stop_losses_percent),
as_percent(self.losing_take_profits_percent)
],
'Percent of total': [
as_percent(self.all_stop_loss_percent),
as_percent(self.all_take_profit_percent)
],
}
df4 = create_summary_table(data4, ["Stop losses", "Take profits"], "Position Exits")
data5 = {
'Biggest realized risk': as_percent(self.max_loss_risk),
'Average realized risk': as_percent(self.avg_realised_risk),
'Max pullback of capital': as_percent(self.max_pullback),
'Sharpe Ratio': as_percent(self.sharpe_ratio),
'Sortino Ratio': as_percent(self.sortino_ratio),
'Profit Factor': as_percent(self.profit_factor),
}
df5 = create_summary_table(data5, "", "Risk Analysis")
display(self.single_column_dfs(df1, df2, df3, df4, df5))
def format_duration(self, duration_timedelta):
if self.time_bucket is not None:
return as_bars(duration_timedelta/self.time_bucket.to_timedelta())
else:
return as_duration(duration_timedelta)
@staticmethod
def single_column_dfs(*dfs):
html = '<div style="display:flex; flex-direction:column">'
for df in dfs:
html += '<div style="margin-bottom: 2em">'
html += df.to_html()
html += '</div>'
html += '</div>'
return HTML(html)
[docs]@dataclass
class TradeAnalysis:
"""Analysis of trades in a portfolio."""
#: The portfolio we analysed
portfolio: Portfolio
#: All taken positions sorted by the position id, or when they were opened
filtered_sorted_positions: list[TradingPosition] = field(init=False)
#: Decision cycle frequency is needed to calculate some performance metrics like sharpe, etc.
#:
#: If not given assume daily for the legacy compatibilty
decision_cycle_frequency: pd.DateOffset = field(default_factory=pd.offsets.Day)
def __post_init__(self):
_filtered_positions = self.portfolio.get_all_positions_filtered()
self.filtered_sorted_positions = sorted(_filtered_positions, key=lambda x: x.position_id)
#assert self.filtered_sorted_positions, "No positions found"
[docs] def get_first_opened_at(self) -> Optional[pd.Timestamp]:
"""Get the opened_at timestamp of the first position in the portfolio."""
return min(
position.opened_at for position in self.filtered_sorted_positions
)
[docs] def get_last_closed_at(self) -> Optional[pd.Timestamp]:
"""Get the closed_at timestamp of the last position in the portfolio."""
return max(
position.closed_at for position in self.filtered_sorted_positions
)
[docs] def get_all_positions(self) -> Iterable[Tuple[PrimaryKey, TradingPosition]]:
"""Return open and closed positions over all traded assets.
Positions are sorted by position_id."""
for position in self.filtered_sorted_positions:
# pair_id, position
yield position.pair.internal_id, position
[docs] def get_open_positions(self) -> Iterable[Tuple[PrimaryKey, TradingPosition]]:
"""Return open positions over all traded assets.
Positions are sorted by position_id."""
for position in self.filtered_sorted_positions:
if position.is_open():
# pair_id, position
yield position.pair.internal_id, position
[docs] def calculate_summary_statistics(
self,
time_bucket: Optional[TimeBucket] = None,
state = None
) -> TradeSummary:
"""Calculate some statistics how our trades went.
:param time_bucket:
Optional, used to display average duration as 'number of bars' instead of 'number of days'.
:param state:
Optional, should be specified if user would like to see advanced statistics
:return:
TradeSummary instance
"""
if(time_bucket is not None):
assert isinstance(time_bucket, TimeBucket), "Not a valid time bucket"
if state is not None:
# for advanced statistics
# import here to avoid circular import error
from tradeexecutor.visual.equity_curve import calculate_daily_returns, calculate_returns, calculate_equity_curve
from tradeexecutor.statistics.key_metric import calculate_sharpe, calculate_sortino, calculate_profit_factor, calculate_max_drawdown, calculate_max_runup
daily_returns = calculate_daily_returns(state, freq="D")
equity_curve = calculate_equity_curve(state)
original_returns = calculate_returns(equity_curve)
sharpe_ratio = calculate_sharpe(daily_returns)
sortino_ratio = calculate_sortino(daily_returns)
# these stats not annualised, so better to calculate it on the original returns
profit_factor = calculate_profit_factor(original_returns)
max_drawdown = calculate_max_drawdown(original_returns)
max_runup = calculate_max_runup(original_returns)
else:
daily_returns = None
original_returns = None
sharpe_ratio = None
sortino_ratio = None
profit_factor = None
max_drawdown = None
max_runup = None
def get_avg_profit_pct_check(trades: List | None):
return float(np.mean(trades)) if trades else None
def get_avg_trade_duration(duration_list: List | None):
if duration_list:
return pd.Timedelta(np.mean(duration_list))
else:
return pd.Timedelta(datetime.timedelta(0))
def func_check(lst, func):
return func(lst) if lst else None
initial_cash = self.portfolio.get_initial_deposit()
uninvested_cash = self.portfolio.get_cash()
# EthLisbon hack
extra_return = 0
duration = datetime.timedelta(0)
winning_trades = []
losing_trades = []
winning_trades_duration = []
losing_trades_duration = []
zero_loss_trades_duration = []
loss_risk_at_open_pc = []
realised_losses = []
biggest_winning_trade_pc = None
biggest_losing_trade_pc = None
average_duration_of_losing_trades = datetime.timedelta(0)
average_duration_of_winning_trades = datetime.timedelta(0)
average_duration_of_zero_loss_trades = None
average_duration_of_all_trades = None
strategy_duration = self.portfolio.get_trading_history_duration()
won = lost = zero_loss = stop_losses = take_profits = undecided = 0
open_value: USDollarAmount = 0
profit: USDollarAmount = 0
trade_volume = 0
lp_fees_paid = 0
max_pos_cons = 0
max_neg_cons = 0
max_pullback_pct = 0
pos_cons = 0
neg_cons = 0
pullback = 0
winning_stop_losses = 0
losing_stop_losses = 0
winning_take_profits = 0
losing_take_profits = 0
for pair_id, position in self.get_all_positions():
portfolio_value_at_open = position.portfolio_value_at_open
capital_tied_at_open_pct = self.get_capital_tied_at_open(position)
if position.stop_loss:
# TODO use maximum_risk
maximum_risk = position.get_loss_risk_at_open()
loss_risk_at_open_pc.append(position.get_loss_risk_at_open_pct())
else:
maximum_risk = None
loss_risk_at_open_pc.append(capital_tied_at_open_pct)
lp_fees_paid += position.get_total_lp_fees_paid() or 0
for t in position.trades.values():
trade_volume += abs(float(t.executed_quantity) * t.executed_price)
if position.is_open():
open_value += position.get_value()
undecided += 1
continue
is_stop_loss = position.is_stop_loss()
is_take_profit = position.is_take_profit()
if is_stop_loss:
stop_losses += 1
if is_take_profit:
take_profits += 1
realised_profit_percent = position.get_realised_profit_percent()
realised_profit_usd = position.get_realised_profit_usd()
duration = position.get_duration()
if position.is_profitable():
won += 1
winning_trades.append(realised_profit_percent)
winning_trades_duration.append(duration)
if is_stop_loss:
winning_stop_losses += 1
if is_take_profit:
winning_take_profits += 1
elif position.is_loss():
lost += 1
losing_trades.append(realised_profit_percent)
losing_trades_duration.append(duration)
if portfolio_value_at_open := position.portfolio_value_at_open:
realised_loss = realised_profit_usd / portfolio_value_at_open
else:
# Bad data
realised_loss = 0
realised_losses.append(realised_loss)
if is_stop_loss:
losing_stop_losses += 1
if is_take_profit:
losing_take_profits += 1
else:
# Any profit exactly balances out loss in slippage and commission
zero_loss += 1
zero_loss_trades_duration.append(duration)
profit += realised_profit_usd
# for getting max consecutive wins/losses and max pullback
# don't do anything if profit = $0
if(realised_profit_usd > 0):
neg_cons = 0
pullback = 0
pos_cons += 1
elif(realised_profit_usd < 0):
pos_cons = 0
neg_cons += 1
pullback += realised_profit_usd
if(neg_cons > max_neg_cons):
max_neg_cons = neg_cons
if(pos_cons > max_pos_cons):
max_pos_cons = pos_cons
if portfolio_value_at_open:
pullback_pct = pullback / (portfolio_value_at_open + realised_profit_usd)
if(pullback_pct < max_pullback_pct):
# pull back is in the negative direction
max_pullback_pct = pullback_pct
else:
# Bad input data / legacy data
max_pullback_pct = 0
all_trades = winning_trades + losing_trades + [0 for i in range(zero_loss)]
average_trade = func_check(all_trades, avg)
median_trade = func_check(all_trades, median)
median_win = func_check(winning_trades, median)
median_loss = func_check(losing_trades, median)
average_winning_trade_profit_pc = get_avg_profit_pct_check(winning_trades)
average_losing_trade_loss_pc = get_avg_profit_pct_check(losing_trades)
max_realised_loss = func_check(realised_losses, min)
avg_realised_risk = func_check(realised_losses, avg)
max_loss_risk_at_open_pc = func_check(loss_risk_at_open_pc, max)
biggest_winning_trade_pc = func_check(winning_trades, max)
biggest_losing_trade_pc = func_check(losing_trades, min)
all_durations = winning_trades_duration + losing_trades_duration + zero_loss_trades_duration
average_duration_of_winning_trades = get_avg_trade_duration(winning_trades_duration)
average_duration_of_losing_trades = get_avg_trade_duration(losing_trades_duration)
if zero_loss_trades_duration:
average_duration_of_zero_loss_trades = get_avg_trade_duration(zero_loss_trades_duration)
if all_durations:
average_duration_of_all_trades = get_avg_trade_duration(all_durations)
lp_fees_average_pc = lp_fees_paid / trade_volume if trade_volume else 0
return TradeSummary(
won=won,
lost=lost,
zero_loss=zero_loss,
stop_losses=stop_losses,
take_profits=take_profits,
undecided=undecided,
realised_profit=profit + extra_return,
open_value=open_value,
uninvested_cash=uninvested_cash,
initial_cash=initial_cash or 0, # Do not pass None for serialisation for live strategies
extra_return=extra_return,
duration=strategy_duration or datetime.timedelta(seconds=0), # Do not pass None for serialisation for live strategies
average_winning_trade_profit_pc=average_winning_trade_profit_pc or 0, # Do not pass None for serialisation for live strategies
average_losing_trade_loss_pc=average_losing_trade_loss_pc or 0, # Do not pass None for serialisation for live strategies
biggest_winning_trade_pc=biggest_winning_trade_pc,
biggest_losing_trade_pc=biggest_losing_trade_pc,
average_duration_of_winning_trades=average_duration_of_winning_trades,
average_duration_of_losing_trades=average_duration_of_losing_trades,
average_duration_of_zero_loss_trades=average_duration_of_zero_loss_trades,
average_duration_of_all_trades=average_duration_of_all_trades,
average_trade=average_trade,
median_trade=median_trade,
median_win=median_win,
median_loss=median_loss,
max_pos_cons=max_pos_cons,
max_neg_cons=max_neg_cons,
max_pullback=max_pullback_pct,
max_drawdown=max_drawdown,
max_runup=max_runup,
max_loss_risk=max_loss_risk_at_open_pc,
max_realised_loss=max_realised_loss,
avg_realised_risk=avg_realised_risk,
time_bucket=time_bucket,
trade_volume=trade_volume,
lp_fees_paid=lp_fees_paid,
lp_fees_average_pc=lp_fees_average_pc,
daily_returns=daily_returns,
winning_stop_losses=winning_stop_losses,
losing_stop_losses=losing_stop_losses,
winning_take_profits=winning_take_profits,
losing_take_profits=losing_take_profits,
sharpe_ratio=sharpe_ratio,
sortino_ratio=sortino_ratio,
profit_factor=profit_factor,
)
[docs] @staticmethod
def get_capital_tied_at_open(position) -> Percent | None:
"""Calculate how much capital % was allocated to this position when it was opened."""
if position.portfolio_value_at_open:
return position.get_capital_tied_at_open_pct()
else:
return None
[docs] def create_timeline(self) -> pd.DataFrame:
"""Create a timeline feed how we traded over a course of time.
Note: We assume each position has only one enter and exit event, not position increases over the lifetime.
:return: DataFrame with timestamp and timeline_event columns
"""
def gen_events():
for pair_id, position in self.get_all_positions():
yield (position.position_id, position)
df = pd.DataFrame(gen_events(), columns=["position_id", "position"])
return df
[docs]class TimelineRowStylingMode(enum.Enum):
#: Style using Pandas background_gradient
gradient = "gradient"
#: Simple
#: Profit = green, loss = red
simple = "simple"
[docs]class TimelineStyler:
"""Style the expanded trades timeline table.
Give HTML hints for DataFrame how it should be rendered
in the notebook output.
"""
[docs] def __init__(self,
row_styling: TimelineRowStylingMode,
hidden_columns: List[str],
vmin: float,
vmax: float,
):
self.row_styling = row_styling
self.hidden_columns = hidden_columns
self.vmin = vmin
self.vmax = vmax
[docs] def colour_timelime_row_simple(self, row: pd.Series) -> pd.Series:
"""Set colour for each timeline row based on its profit.
- +/- 5% colouring
- More information: https://stackoverflow.com/a/49745352/315168
- CSS colours: https://htmlcolorcodes.com/color-names/
"""
pnl_raw = row["PnL % raw"]
if pnl_raw < -0.05:
return pd.Series('background-color: Salmon', row.index)
elif pnl_raw < 0:
return pd.Series('background-color: LightSalmon', row.index)
elif pnl_raw > 0.05:
return pd.Series('background-color: LawnGreen', row.index)
else:
return pd.Series('background-color: PaleGreen', row.index)
[docs] def __call__(self, df: pd.DataFrame):
"""Applies styles on a dataframe
:param df:
Dataframe as returned by :py:func`expand_timeline`.
"""
# Create a Pandas Styler with multiple styling options applied
try:
styles = df.style \
.hide(axis="index") \
.hide(axis="columns", subset=self.hidden_columns)
except KeyError:
# The input df was empty (no trades)
styles = df.style
# Don't let the text inside a cell to wrap
styles = styles.set_table_styles({
"Opened at": [{'selector': 'td', 'props': [('white-space', 'nowrap')]}],
"Exchange": [{'selector': 'td', 'props': [('white-space', 'nowrap')]}],
})
if self.row_styling == TimelineRowStylingMode.gradient:
# Dynamically color the background of trade outcome coluns # https://pandas.pydata.org/docs/reference/api/pandas.io.formats.style.Styler.background_gradient.html
# TODO: This gradient styling is confusing
# get rid of it long term
styles = styles.background_gradient(
axis=0,
gmap=df['PnL % raw'],
cmap='RdYlGn',
vmin=self.vmin, # We can only lose 100% of our money on position
vmax=self.vmax) # 50% profit is 21.5 position. Assume this is the max success color we can hit over
else:
styles = styles.apply(self.colour_timelime_row_simple, axis=1)
return styles
[docs]def expand_timeline(
exchanges: Set[Exchange],
pair_universe: PandasPairUniverse,
timeline: pd.DataFrame,
vmin=-0.3,
vmax=0.2,
timestamp_format="%Y-%m-%d",
hidden_columns=["Id", "PnL % raw"],
row_styling_mode=TimelineRowStylingMode.simple,
) -> Tuple[pd.DataFrame, TimelineStyler]:
"""Expand trade history timeline to human readable table.
This will the outputting much easier in Python Notebooks.
Currently does not incrementing/decreasing positions gradually.
Instaqd of applying styles or returning a styled dataframe, we return a callable that applies the styles.
This is because of Pandas issue https://github.com/pandas-dev/pandas/issues/40675 - hidden indexes, columns,
etc. are not exported.
:param exchanges: Needed for exchange metadata
:param pair_universe: Needed for trading pair metadata
:param vmax: Trade success % to have the extreme green color.
:param vmin: The % of lost capital on the trade to have the extreme red color.
:param timestamp_format: How to format Opened at column, as passed to `strftime()`
:param hidden_columns: Hide columns in the output table
:return: DataFrame with human=readable position win/loss information, having DF indexed by timestamps and a styler function
"""
exchange_map = {e.exchange_id: e for e in exchanges}
# https://stackoverflow.com/a/52363890/315168
def expander(row):
position: TradingPosition = row["position"]
# timestamp = row.name # ???
pair_id = position.pair.internal_id
pair_info = pair_universe.get_pair_by_id(pair_id)
exchange = exchange_map.get(pair_info.exchange_id)
if not exchange:
raise RuntimeError(f"No exchange for id {pair_info.exchange_id}, pair {pair_info}")
if position.is_stop_loss():
remarks = "SL"
elif position.is_take_profit():
remarks = "TP"
else:
remarks = ""
# Hack around to work with legacy data issue.
# Not an issue for new strategies.
if position.has_bad_data_issues():
remarks += "BAD"
duration = position.get_duration()
r = {
# "timestamp": timestamp,
"Id": position.position_id,
"Remarks": remarks,
"Opened at": position.opened_at.strftime(timestamp_format),
"Duration": format_duration_days_hours_mins(duration) if duration else np.nan,
"Exchange": exchange.name,
"Base asset": pair_info.base_token_symbol,
"Quote asset": pair_info.quote_token_symbol,
"Position max value": format_value(position.get_max_size()),
"PnL USD": format_value(position.get_realised_profit_usd()) if position.is_closed() else np.nan,
"PnL %": format_percent_2_decimals(position.get_realised_profit_percent()) if position.is_closed() else np.nan,
"PnL % raw": position.get_realised_profit_percent() if position.is_closed() else 0,
"Open mid price USD": format_price(position.get_opening_price()),
"Close mid price USD": format_price(position.get_closing_price()) if position.is_closed() else np.nan,
"Trade count": position.get_trade_count(),
"LP fees": f"${position.get_total_lp_fees_paid():,.2f}"
}
return r
applied_df = timeline.apply(expander, axis='columns', result_type='expand')
if len(applied_df) > 0:
# https://stackoverflow.com/a/52720936/315168
applied_df \
.sort_values(by=['Id'], ascending=[True], inplace=True)
# Get rid of NaN labels
# https://stackoverflow.com/a/28390992/315168
applied_df.fillna('', inplace=True)
styling = TimelineStyler(
row_styling=row_styling_mode,
hidden_columns=hidden_columns,
vmin=vmin,
vmax=vmax,
)
return applied_df, styling
[docs]def expand_timeline_raw(
timeline: pd.DataFrame,
timestamp_format="%Y-%m-%d"
) -> pd.DataFrame:
"""A simplified version of expand_timeline that does not care about
pair info, exchanges, or opening capital, and also provides raw figures.
Unused in codebase, but can be useful for advanced users to use directly"""
# https://stackoverflow.com/a/52363890/315168
def expander(row):
position: TradingPosition = row["position"]
# timestamp = row.name # ???
pair_id = position.pair.internal_id
if position.is_stop_loss():
remarks = "SL"
elif position.is_take_profit():
remarks = "TP"
else:
remarks = ""
pnl_usd = position.get_realised_profit_usd() if position.is_closed() else np.nan
duration = position.get_duration()
r = {
# "timestamp": timestamp,
"Id": position.position_id,
"Remarks": remarks,
"Opened at": position.opened_at.strftime(timestamp_format),
"Duration": format_duration_days_hours_mins(duration) if duration else np.nan,
"position_max_size": position.get_max_size(),
"pnl_usd": pnl_usd,
"pnl_pct_raw": position.get_realised_profit_percent() if position.is_closed() else 0,
"open_price_usd": position.get_opening_price(),
"close_price_usd": position.get_closing_price() if position.is_closed() else np.nan,
"trade_count": position.get_trade_count(),
}
return r
applied_df = timeline.apply(expander, axis='columns', result_type='expand')
if len(applied_df) > 0:
# https://stackoverflow.com/a/52720936/315168
applied_df \
.sort_values(by=['Id'], ascending=[True], inplace=True)
# Get rid of NaN labels
# https://stackoverflow.com/a/28390992/315168
applied_df.fillna('', inplace=True)
return applied_df
[docs]def build_trade_analysis(
portfolio: Portfolio
) -> TradeAnalysis:
"""Build a trade analysis from list of positions.
- Read positions from backtesting or live state
- Create TradeAnalysis instance that can be used to display IPython notebook
data on the performance
"""
return TradeAnalysis(
portfolio,
)
[docs]def avg(lst: list[int]):
return sum(lst) / len(lst)
[docs]def avg_timedelta(lst: list[pd.Timedelta]):
return pd.Series(lst).mean()