mirror of
https://github.com/freqtrade/freqtrade.git
synced 2024-11-10 10:21:59 +00:00
718 lines
25 KiB
Python
718 lines
25 KiB
Python
import logging
|
|
from datetime import datetime, timezone
|
|
from pathlib import Path
|
|
from typing import Dict, List, Optional
|
|
|
|
import pandas as pd
|
|
|
|
from freqtrade.configuration import TimeRange
|
|
from freqtrade.constants import Config
|
|
from freqtrade.data.btanalysis import (
|
|
analyze_trade_parallelism,
|
|
extract_trades_of_period,
|
|
load_trades,
|
|
)
|
|
from freqtrade.data.converter import trim_dataframe
|
|
from freqtrade.data.dataprovider import DataProvider
|
|
from freqtrade.data.history import get_timerange, load_data
|
|
from freqtrade.data.metrics import (
|
|
calculate_max_drawdown,
|
|
calculate_underwater,
|
|
combine_dataframes_with_mean,
|
|
create_cum_profit,
|
|
)
|
|
from freqtrade.enums import CandleType
|
|
from freqtrade.exceptions import OperationalException
|
|
from freqtrade.exchange import timeframe_to_prev_date, timeframe_to_seconds
|
|
from freqtrade.misc import pair_to_filename
|
|
from freqtrade.plugins.pairlist.pairlist_helpers import expand_pairlist
|
|
from freqtrade.resolvers import ExchangeResolver, StrategyResolver
|
|
from freqtrade.strategy import IStrategy
|
|
from freqtrade.strategy.strategy_wrapper import strategy_safe_wrapper
|
|
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
|
|
try:
|
|
import plotly.graph_objects as go
|
|
from plotly.offline import plot
|
|
from plotly.subplots import make_subplots
|
|
except ImportError:
|
|
logger.exception("Module plotly not found \n Please install using `pip3 install plotly`")
|
|
exit(1)
|
|
|
|
|
|
def init_plotscript(config, markets: List, startup_candles: int = 0):
|
|
"""
|
|
Initialize objects needed for plotting
|
|
:return: Dict with candle (OHLCV) data, trades and pairs
|
|
"""
|
|
|
|
if "pairs" in config:
|
|
pairs = expand_pairlist(config["pairs"], markets)
|
|
else:
|
|
pairs = expand_pairlist(config["exchange"]["pair_whitelist"], markets)
|
|
|
|
# Set timerange to use
|
|
timerange = TimeRange.parse_timerange(config.get("timerange"))
|
|
|
|
data = load_data(
|
|
datadir=config.get("datadir"),
|
|
pairs=pairs,
|
|
timeframe=config["timeframe"],
|
|
timerange=timerange,
|
|
startup_candles=startup_candles,
|
|
data_format=config["dataformat_ohlcv"],
|
|
candle_type=config.get("candle_type_def", CandleType.SPOT),
|
|
)
|
|
|
|
if startup_candles and data:
|
|
min_date, max_date = get_timerange(data)
|
|
logger.info(f"Loading data from {min_date} to {max_date}")
|
|
timerange.adjust_start_if_necessary(
|
|
timeframe_to_seconds(config["timeframe"]), startup_candles, min_date
|
|
)
|
|
|
|
no_trades = False
|
|
filename = config.get("exportfilename")
|
|
if config.get("no_trades", False):
|
|
no_trades = True
|
|
elif config["trade_source"] == "file":
|
|
if not filename.is_dir() and not filename.is_file():
|
|
logger.warning("Backtest file is missing skipping trades.")
|
|
no_trades = True
|
|
try:
|
|
trades = load_trades(
|
|
config["trade_source"],
|
|
db_url=config.get("db_url"),
|
|
exportfilename=filename,
|
|
no_trades=no_trades,
|
|
strategy=config.get("strategy"),
|
|
)
|
|
except ValueError as e:
|
|
raise OperationalException(e) from e
|
|
if not trades.empty:
|
|
trades = trim_dataframe(trades, timerange, df_date_col="open_date")
|
|
|
|
return {
|
|
"ohlcv": data,
|
|
"trades": trades,
|
|
"pairs": pairs,
|
|
"timerange": timerange,
|
|
}
|
|
|
|
|
|
def add_indicators(fig, row, indicators: Dict[str, Dict], data: pd.DataFrame) -> make_subplots:
|
|
"""
|
|
Generate all the indicators selected by the user for a specific row, based on the configuration
|
|
:param fig: Plot figure to append to
|
|
:param row: row number for this plot
|
|
:param indicators: Dict of Indicators with configuration options.
|
|
Dict key must correspond to dataframe column.
|
|
:param data: candlestick DataFrame
|
|
"""
|
|
plot_kinds = {
|
|
"scatter": go.Scatter,
|
|
"bar": go.Bar,
|
|
}
|
|
for indicator, conf in indicators.items():
|
|
logger.debug(f"indicator {indicator} with config {conf}")
|
|
if indicator in data:
|
|
kwargs = {"x": data["date"], "y": data[indicator].values, "name": indicator}
|
|
|
|
plot_type = conf.get("type", "scatter")
|
|
color = conf.get("color")
|
|
if plot_type == "bar":
|
|
kwargs.update(
|
|
{
|
|
"marker_color": color or "DarkSlateGrey",
|
|
"marker_line_color": color or "DarkSlateGrey",
|
|
}
|
|
)
|
|
else:
|
|
if color:
|
|
kwargs.update({"line": {"color": color}})
|
|
kwargs["mode"] = "lines"
|
|
if plot_type != "scatter":
|
|
logger.warning(
|
|
f"Indicator {indicator} has unknown plot trace kind {plot_type}"
|
|
f', assuming "scatter".'
|
|
)
|
|
|
|
kwargs.update(conf.get("plotly", {}))
|
|
trace = plot_kinds[plot_type](**kwargs)
|
|
fig.add_trace(trace, row, 1)
|
|
else:
|
|
logger.info(
|
|
'Indicator "%s" ignored. Reason: This indicator is not found ' "in your strategy.",
|
|
indicator,
|
|
)
|
|
|
|
return fig
|
|
|
|
|
|
def add_profit(fig, row, data: pd.DataFrame, column: str, name: str) -> make_subplots:
|
|
"""
|
|
Add profit-plot
|
|
:param fig: Plot figure to append to
|
|
:param row: row number for this plot
|
|
:param data: candlestick DataFrame
|
|
:param column: Column to use for plot
|
|
:param name: Name to use
|
|
:return: fig with added profit plot
|
|
"""
|
|
profit = go.Scatter(
|
|
x=data.index,
|
|
y=data[column],
|
|
name=name,
|
|
)
|
|
fig.add_trace(profit, row, 1)
|
|
|
|
return fig
|
|
|
|
|
|
def add_max_drawdown(
|
|
fig, row, trades: pd.DataFrame, df_comb: pd.DataFrame, timeframe: str, starting_balance: float
|
|
) -> make_subplots:
|
|
"""
|
|
Add scatter points indicating max drawdown
|
|
"""
|
|
try:
|
|
drawdown = calculate_max_drawdown(trades, starting_balance=starting_balance)
|
|
|
|
drawdown = go.Scatter(
|
|
x=[drawdown.high_date, drawdown.low_date],
|
|
y=[
|
|
df_comb.loc[timeframe_to_prev_date(timeframe, drawdown.high_date), "cum_profit"],
|
|
df_comb.loc[timeframe_to_prev_date(timeframe, drawdown.low_date), "cum_profit"],
|
|
],
|
|
mode="markers",
|
|
name=f"Max drawdown {drawdown.relative_account_drawdown:.2%}",
|
|
text=f"Max drawdown {drawdown.relative_account_drawdown:.2%}",
|
|
marker=dict(symbol="square-open", size=9, line=dict(width=2), color="green"),
|
|
)
|
|
fig.add_trace(drawdown, row, 1)
|
|
except ValueError:
|
|
logger.warning("No trades found - not plotting max drawdown.")
|
|
return fig
|
|
|
|
|
|
def add_underwater(fig, row, trades: pd.DataFrame, starting_balance: float) -> make_subplots:
|
|
"""
|
|
Add underwater plots
|
|
"""
|
|
try:
|
|
underwater = calculate_underwater(
|
|
trades, value_col="profit_abs", starting_balance=starting_balance
|
|
)
|
|
|
|
underwater_plot = go.Scatter(
|
|
x=underwater["date"],
|
|
y=underwater["drawdown"],
|
|
name="Underwater Plot",
|
|
fill="tozeroy",
|
|
fillcolor="#cc362b",
|
|
line={"color": "#cc362b"},
|
|
)
|
|
|
|
underwater_plot_relative = go.Scatter(
|
|
x=underwater["date"],
|
|
y=(-underwater["drawdown_relative"]),
|
|
name="Underwater Plot (%)",
|
|
fill="tozeroy",
|
|
fillcolor="green",
|
|
line={"color": "green"},
|
|
)
|
|
|
|
fig.add_trace(underwater_plot, row, 1)
|
|
fig.add_trace(underwater_plot_relative, row + 1, 1)
|
|
except ValueError:
|
|
logger.warning("No trades found - not plotting underwater plot")
|
|
return fig
|
|
|
|
|
|
def add_parallelism(fig, row, trades: pd.DataFrame, timeframe: str) -> make_subplots:
|
|
"""
|
|
Add Chart showing trade parallelism
|
|
"""
|
|
try:
|
|
result = analyze_trade_parallelism(trades, timeframe)
|
|
|
|
drawdown = go.Scatter(
|
|
x=result.index,
|
|
y=result["open_trades"],
|
|
name="Parallel trades",
|
|
fill="tozeroy",
|
|
fillcolor="#242222",
|
|
line={"color": "#242222"},
|
|
)
|
|
fig.add_trace(drawdown, row, 1)
|
|
except ValueError:
|
|
logger.warning("No trades found - not plotting Parallelism.")
|
|
return fig
|
|
|
|
|
|
def plot_trades(fig, trades: pd.DataFrame) -> make_subplots:
|
|
"""
|
|
Add trades to "fig"
|
|
"""
|
|
# Trades can be empty
|
|
if trades is not None and len(trades) > 0:
|
|
# Create description for exit summarizing the trade
|
|
trades["desc"] = trades.apply(
|
|
lambda row: f"{row['profit_ratio']:.2%}, "
|
|
+ (f"{row['enter_tag']}, " if row["enter_tag"] is not None else "")
|
|
+ f"{row['exit_reason']}, "
|
|
+ f"{row['trade_duration']} min",
|
|
axis=1,
|
|
)
|
|
trade_entries = go.Scatter(
|
|
x=trades["open_date"],
|
|
y=trades["open_rate"],
|
|
mode="markers",
|
|
name="Trade entry",
|
|
text=trades["desc"],
|
|
marker=dict(symbol="circle-open", size=11, line=dict(width=2), color="cyan"),
|
|
)
|
|
|
|
trade_exits = go.Scatter(
|
|
x=trades.loc[trades["profit_ratio"] > 0, "close_date"],
|
|
y=trades.loc[trades["profit_ratio"] > 0, "close_rate"],
|
|
text=trades.loc[trades["profit_ratio"] > 0, "desc"],
|
|
mode="markers",
|
|
name="Exit - Profit",
|
|
marker=dict(symbol="square-open", size=11, line=dict(width=2), color="green"),
|
|
)
|
|
trade_exits_loss = go.Scatter(
|
|
x=trades.loc[trades["profit_ratio"] <= 0, "close_date"],
|
|
y=trades.loc[trades["profit_ratio"] <= 0, "close_rate"],
|
|
text=trades.loc[trades["profit_ratio"] <= 0, "desc"],
|
|
mode="markers",
|
|
name="Exit - Loss",
|
|
marker=dict(symbol="square-open", size=11, line=dict(width=2), color="red"),
|
|
)
|
|
fig.add_trace(trade_entries, 1, 1)
|
|
fig.add_trace(trade_exits, 1, 1)
|
|
fig.add_trace(trade_exits_loss, 1, 1)
|
|
else:
|
|
logger.warning("No trades found.")
|
|
return fig
|
|
|
|
|
|
def create_plotconfig(
|
|
indicators1: List[str], indicators2: List[str], plot_config: Dict[str, Dict]
|
|
) -> Dict[str, Dict]:
|
|
"""
|
|
Combines indicators 1 and indicators 2 into plot_config if necessary
|
|
:param indicators1: List containing Main plot indicators
|
|
:param indicators2: List containing Sub plot indicators
|
|
:param plot_config: Dict of Dicts containing advanced plot configuration
|
|
:return: plot_config - eventually with indicators 1 and 2
|
|
"""
|
|
|
|
if plot_config:
|
|
if indicators1:
|
|
plot_config["main_plot"] = {ind: {} for ind in indicators1}
|
|
if indicators2:
|
|
plot_config["subplots"] = {"Other": {ind: {} for ind in indicators2}}
|
|
|
|
if not plot_config:
|
|
# If no indicators and no plot-config given, use defaults.
|
|
if not indicators1:
|
|
indicators1 = ["sma", "ema3", "ema5"]
|
|
if not indicators2:
|
|
indicators2 = ["macd", "macdsignal"]
|
|
|
|
# Create subplot configuration if plot_config is not available.
|
|
plot_config = {
|
|
"main_plot": {ind: {} for ind in indicators1},
|
|
"subplots": {"Other": {ind: {} for ind in indicators2}},
|
|
}
|
|
if "main_plot" not in plot_config:
|
|
plot_config["main_plot"] = {}
|
|
|
|
if "subplots" not in plot_config:
|
|
plot_config["subplots"] = {}
|
|
return plot_config
|
|
|
|
|
|
def plot_area(
|
|
fig,
|
|
row: int,
|
|
data: pd.DataFrame,
|
|
indicator_a: str,
|
|
indicator_b: str,
|
|
label: str = "",
|
|
fill_color: str = "rgba(0,176,246,0.2)",
|
|
) -> make_subplots:
|
|
"""Creates a plot for the area between two traces and adds it to fig.
|
|
:param fig: Plot figure to append to
|
|
:param row: row number for this plot
|
|
:param data: candlestick DataFrame
|
|
:param indicator_a: indicator name as populated in strategy
|
|
:param indicator_b: indicator name as populated in strategy
|
|
:param label: label for the filled area
|
|
:param fill_color: color to be used for the filled area
|
|
:return: fig with added filled_traces plot
|
|
"""
|
|
if indicator_a in data and indicator_b in data:
|
|
# make lines invisible to get the area plotted, only.
|
|
line = {"color": "rgba(255,255,255,0)"}
|
|
# TODO: Figure out why scattergl causes problems plotly/plotly.js#2284
|
|
trace_a = go.Scatter(x=data.date, y=data[indicator_a], showlegend=False, line=line)
|
|
trace_b = go.Scatter(
|
|
x=data.date,
|
|
y=data[indicator_b],
|
|
name=label,
|
|
fill="tonexty",
|
|
fillcolor=fill_color,
|
|
line=line,
|
|
)
|
|
fig.add_trace(trace_a, row, 1)
|
|
fig.add_trace(trace_b, row, 1)
|
|
return fig
|
|
|
|
|
|
def add_areas(fig, row: int, data: pd.DataFrame, indicators) -> make_subplots:
|
|
"""Adds all area plots (specified in plot_config) to fig.
|
|
:param fig: Plot figure to append to
|
|
:param row: row number for this plot
|
|
:param data: candlestick DataFrame
|
|
:param indicators: dict with indicators. ie.: plot_config['main_plot'] or
|
|
plot_config['subplots'][subplot_label]
|
|
:return: fig with added filled_traces plot
|
|
"""
|
|
for indicator, ind_conf in indicators.items():
|
|
if "fill_to" in ind_conf:
|
|
indicator_b = ind_conf["fill_to"]
|
|
if indicator in data and indicator_b in data:
|
|
label = ind_conf.get("fill_label", f"{indicator}<>{indicator_b}")
|
|
fill_color = ind_conf.get("fill_color", "rgba(0,176,246,0.2)")
|
|
fig = plot_area(
|
|
fig, row, data, indicator, indicator_b, label=label, fill_color=fill_color
|
|
)
|
|
elif indicator not in data:
|
|
logger.info(
|
|
'Indicator "%s" ignored. Reason: This indicator is not '
|
|
"found in your strategy.",
|
|
indicator,
|
|
)
|
|
elif indicator_b not in data:
|
|
logger.info(
|
|
'fill_to: "%s" ignored. Reason: This indicator is not ' "in your strategy.",
|
|
indicator_b,
|
|
)
|
|
return fig
|
|
|
|
|
|
def create_scatter(data, column_name, color, direction) -> Optional[go.Scatter]:
|
|
if column_name in data.columns:
|
|
df_short = data[data[column_name] == 1]
|
|
if len(df_short) > 0:
|
|
shorts = go.Scatter(
|
|
x=df_short.date,
|
|
y=df_short.close,
|
|
mode="markers",
|
|
name=column_name,
|
|
marker=dict(
|
|
symbol=f"triangle-{direction}-dot",
|
|
size=9,
|
|
line=dict(width=1),
|
|
color=color,
|
|
),
|
|
)
|
|
return shorts
|
|
else:
|
|
logger.warning(f"No {column_name}-signals found.")
|
|
|
|
return None
|
|
|
|
|
|
def generate_candlestick_graph(
|
|
pair: str,
|
|
data: pd.DataFrame,
|
|
trades: Optional[pd.DataFrame] = None,
|
|
*,
|
|
indicators1: Optional[List[str]] = None,
|
|
indicators2: Optional[List[str]] = None,
|
|
plot_config: Optional[Dict[str, Dict]] = None,
|
|
) -> go.Figure:
|
|
"""
|
|
Generate the graph from the data generated by Backtesting or from DB
|
|
Volume will always be plotted in row2, so Row 1 and 3 are to our disposal for custom indicators
|
|
:param pair: Pair to Display on the graph
|
|
:param data: OHLCV DataFrame containing indicators and entry/exit signals
|
|
:param trades: All trades created
|
|
:param indicators1: List containing Main plot indicators
|
|
:param indicators2: List containing Sub plot indicators
|
|
:param plot_config: Dict of Dicts containing advanced plot configuration
|
|
:return: Plotly figure
|
|
"""
|
|
plot_config = create_plotconfig(
|
|
indicators1 or [],
|
|
indicators2 or [],
|
|
plot_config or {},
|
|
)
|
|
rows = 2 + len(plot_config["subplots"])
|
|
row_widths = [1 for _ in plot_config["subplots"]]
|
|
# Define the graph
|
|
fig = make_subplots(
|
|
rows=rows,
|
|
cols=1,
|
|
shared_xaxes=True,
|
|
row_width=row_widths + [1, 4],
|
|
vertical_spacing=0.0001,
|
|
)
|
|
fig["layout"].update(title=pair)
|
|
fig["layout"]["yaxis1"].update(title="Price")
|
|
fig["layout"]["yaxis2"].update(title="Volume")
|
|
for i, name in enumerate(plot_config["subplots"]):
|
|
fig["layout"][f"yaxis{3 + i}"].update(title=name)
|
|
fig["layout"]["xaxis"]["rangeslider"].update(visible=False)
|
|
fig.update_layout(modebar_add=["v1hovermode", "toggleSpikeLines"])
|
|
|
|
# Common information
|
|
candles = go.Candlestick(
|
|
x=data.date, open=data.open, high=data.high, low=data.low, close=data.close, name="Price"
|
|
)
|
|
fig.add_trace(candles, 1, 1)
|
|
|
|
longs = create_scatter(data, "enter_long", "green", "up")
|
|
exit_longs = create_scatter(data, "exit_long", "red", "down")
|
|
shorts = create_scatter(data, "enter_short", "blue", "down")
|
|
exit_shorts = create_scatter(data, "exit_short", "violet", "up")
|
|
|
|
for scatter in [longs, exit_longs, shorts, exit_shorts]:
|
|
if scatter:
|
|
fig.add_trace(scatter, 1, 1)
|
|
|
|
# Add Bollinger Bands
|
|
fig = plot_area(fig, 1, data, "bb_lowerband", "bb_upperband", label="Bollinger Band")
|
|
# prevent bb_lower and bb_upper from plotting
|
|
try:
|
|
del plot_config["main_plot"]["bb_lowerband"]
|
|
del plot_config["main_plot"]["bb_upperband"]
|
|
except KeyError:
|
|
pass
|
|
# main plot goes to row 1
|
|
fig = add_indicators(fig=fig, row=1, indicators=plot_config["main_plot"], data=data)
|
|
fig = add_areas(fig, 1, data, plot_config["main_plot"])
|
|
fig = plot_trades(fig, trades)
|
|
# sub plot: Volume goes to row 2
|
|
volume = go.Bar(
|
|
x=data["date"],
|
|
y=data["volume"],
|
|
name="Volume",
|
|
marker_color="DarkSlateGrey",
|
|
marker_line_color="DarkSlateGrey",
|
|
)
|
|
fig.add_trace(volume, 2, 1)
|
|
# add each sub plot to a separate row
|
|
for i, label in enumerate(plot_config["subplots"]):
|
|
sub_config = plot_config["subplots"][label]
|
|
row = 3 + i
|
|
fig = add_indicators(fig=fig, row=row, indicators=sub_config, data=data)
|
|
# fill area between indicators ( 'fill_to': 'other_indicator')
|
|
fig = add_areas(fig, row, data, sub_config)
|
|
|
|
return fig
|
|
|
|
|
|
def generate_profit_graph(
|
|
pairs: str,
|
|
data: Dict[str, pd.DataFrame],
|
|
trades: pd.DataFrame,
|
|
timeframe: str,
|
|
stake_currency: str,
|
|
starting_balance: float,
|
|
) -> go.Figure:
|
|
# Combine close-values for all pairs, rename columns to "pair"
|
|
try:
|
|
df_comb = combine_dataframes_with_mean(data, "close")
|
|
except ValueError:
|
|
raise OperationalException(
|
|
"No data found. Please make sure that data is available for "
|
|
"the timerange and pairs selected."
|
|
)
|
|
|
|
# Trim trades to available OHLCV data
|
|
trades = extract_trades_of_period(df_comb, trades, date_index=True)
|
|
if len(trades) == 0:
|
|
raise OperationalException("No trades found in selected timerange.")
|
|
|
|
# Add combined cumulative profit
|
|
df_comb = create_cum_profit(df_comb, trades, "cum_profit", timeframe)
|
|
|
|
# Plot the pairs average close prices, and total profit growth
|
|
avgclose = go.Scatter(
|
|
x=df_comb.index,
|
|
y=df_comb["mean"],
|
|
name="Avg close price",
|
|
)
|
|
|
|
fig = make_subplots(
|
|
rows=6,
|
|
cols=1,
|
|
shared_xaxes=True,
|
|
row_heights=[1, 1, 1, 0.5, 0.75, 0.75],
|
|
vertical_spacing=0.05,
|
|
subplot_titles=[
|
|
"AVG Close Price",
|
|
"Combined Profit",
|
|
"Profit per pair",
|
|
"Parallelism",
|
|
"Underwater",
|
|
"Relative Drawdown",
|
|
],
|
|
)
|
|
fig["layout"].update(title="Freqtrade Profit plot")
|
|
fig["layout"]["yaxis1"].update(title="Price")
|
|
fig["layout"]["yaxis2"].update(title=f"Profit {stake_currency}")
|
|
fig["layout"]["yaxis3"].update(title=f"Profit {stake_currency}")
|
|
fig["layout"]["yaxis4"].update(title="Trade count")
|
|
fig["layout"]["yaxis5"].update(title="Underwater Plot")
|
|
fig["layout"]["yaxis6"].update(title="Underwater Plot Relative (%)", tickformat=",.2%")
|
|
fig["layout"]["xaxis"]["rangeslider"].update(visible=False)
|
|
fig.update_layout(modebar_add=["v1hovermode", "toggleSpikeLines"])
|
|
|
|
fig.add_trace(avgclose, 1, 1)
|
|
fig = add_profit(fig, 2, df_comb, "cum_profit", "Profit")
|
|
fig = add_max_drawdown(fig, 2, trades, df_comb, timeframe, starting_balance)
|
|
fig = add_parallelism(fig, 4, trades, timeframe)
|
|
# Two rows consumed
|
|
fig = add_underwater(fig, 5, trades, starting_balance)
|
|
|
|
for pair in pairs:
|
|
profit_col = f"cum_profit_{pair}"
|
|
try:
|
|
df_comb = create_cum_profit(
|
|
df_comb, trades[trades["pair"] == pair], profit_col, timeframe
|
|
)
|
|
fig = add_profit(fig, 3, df_comb, profit_col, f"Profit {pair}")
|
|
except ValueError:
|
|
pass
|
|
return fig
|
|
|
|
|
|
def generate_plot_filename(pair: str, timeframe: str) -> str:
|
|
"""
|
|
Generate filenames per pair/timeframe to be used for storing plots
|
|
"""
|
|
pair_s = pair_to_filename(pair)
|
|
file_name = "freqtrade-plot-" + pair_s + "-" + timeframe + ".html"
|
|
|
|
logger.info("Generate plot file for %s", pair)
|
|
|
|
return file_name
|
|
|
|
|
|
def store_plot_file(fig, filename: str, directory: Path, auto_open: bool = False) -> None:
|
|
"""
|
|
Generate a plot html file from pre populated fig plotly object
|
|
:param fig: Plotly Figure to plot
|
|
:param filename: Name to store the file as
|
|
:param directory: Directory to store the file in
|
|
:param auto_open: Automatically open files saved
|
|
:return: None
|
|
"""
|
|
directory.mkdir(parents=True, exist_ok=True)
|
|
|
|
_filename = directory.joinpath(filename)
|
|
plot(fig, filename=str(_filename), auto_open=auto_open)
|
|
logger.info(f"Stored plot as {_filename}")
|
|
|
|
|
|
def load_and_plot_trades(config: Config):
|
|
"""
|
|
From configuration provided
|
|
- Initializes plot-script
|
|
- Get candle (OHLCV) data
|
|
- Generate Dafaframes populated with indicators and signals based on configured strategy
|
|
- Load trades executed during the selected period
|
|
- Generate Plotly plot objects
|
|
- Generate plot files
|
|
:return: None
|
|
"""
|
|
strategy = StrategyResolver.load_strategy(config)
|
|
|
|
exchange = ExchangeResolver.load_exchange(config)
|
|
IStrategy.dp = DataProvider(config, exchange)
|
|
strategy.ft_bot_start()
|
|
strategy_safe_wrapper(strategy.bot_loop_start)(current_time=datetime.now(timezone.utc))
|
|
plot_elements = init_plotscript(config, list(exchange.markets), strategy.startup_candle_count)
|
|
timerange = plot_elements["timerange"]
|
|
trades = plot_elements["trades"]
|
|
pair_counter = 0
|
|
for pair, data in plot_elements["ohlcv"].items():
|
|
pair_counter += 1
|
|
logger.info("analyse pair %s", pair)
|
|
|
|
df_analyzed = strategy.analyze_ticker(data, {"pair": pair})
|
|
df_analyzed = trim_dataframe(df_analyzed, timerange)
|
|
if not trades.empty:
|
|
trades_pair = trades.loc[trades["pair"] == pair]
|
|
trades_pair = extract_trades_of_period(df_analyzed, trades_pair)
|
|
else:
|
|
trades_pair = trades
|
|
|
|
fig = generate_candlestick_graph(
|
|
pair=pair,
|
|
data=df_analyzed,
|
|
trades=trades_pair,
|
|
indicators1=config.get("indicators1", []),
|
|
indicators2=config.get("indicators2", []),
|
|
plot_config=strategy.plot_config if hasattr(strategy, "plot_config") else {},
|
|
)
|
|
|
|
store_plot_file(
|
|
fig,
|
|
filename=generate_plot_filename(pair, config["timeframe"]),
|
|
directory=config["user_data_dir"] / "plot",
|
|
)
|
|
|
|
logger.info("End of plotting process. %s plots generated", pair_counter)
|
|
|
|
|
|
def plot_profit(config: Config) -> None:
|
|
"""
|
|
Plots the total profit for all pairs.
|
|
Note, the profit calculation isn't realistic.
|
|
But should be somewhat proportional, and therefore useful
|
|
in helping out to find a good algorithm.
|
|
"""
|
|
if "timeframe" not in config:
|
|
raise OperationalException("Timeframe must be set in either config or via --timeframe.")
|
|
|
|
exchange = ExchangeResolver.load_exchange(config)
|
|
plot_elements = init_plotscript(config, list(exchange.markets))
|
|
trades = plot_elements["trades"]
|
|
# Filter trades to relevant pairs
|
|
# Remove open pairs - we don't know the profit yet so can't calculate profit for these.
|
|
# Also, If only one open pair is left, then the profit-generation would fail.
|
|
trades = trades[
|
|
(trades["pair"].isin(plot_elements["pairs"])) & (~trades["close_date"].isnull())
|
|
]
|
|
if len(trades) == 0:
|
|
raise OperationalException(
|
|
"No trades found, cannot generate Profit-plot without "
|
|
"trades from either Backtest result or database."
|
|
)
|
|
|
|
# Create an average close price of all the pairs that were involved.
|
|
# this could be useful to gauge the overall market trend
|
|
fig = generate_profit_graph(
|
|
plot_elements["pairs"],
|
|
plot_elements["ohlcv"],
|
|
trades,
|
|
config["timeframe"],
|
|
config.get("stake_currency", ""),
|
|
config.get("available_capital", config["dry_run_wallet"]),
|
|
)
|
|
store_plot_file(
|
|
fig,
|
|
filename="freqtrade-profit-plot.html",
|
|
directory=config["user_data_dir"] / "plot",
|
|
auto_open=config.get("plot_auto_open", False),
|
|
)
|