stable/tests/optimize/test_optimize_reports.py
2022-04-11 18:02:06 +02:00

405 lines
17 KiB
Python

import re
from datetime import timedelta
from pathlib import Path
import pandas as pd
import pytest
from arrow import Arrow
from freqtrade.configuration import TimeRange
from freqtrade.constants import DATETIME_PRINT_FORMAT, LAST_BT_RESULT_FN
from freqtrade.data import history
from freqtrade.data.btanalysis import (get_latest_backtest_filename, load_backtest_data,
load_backtest_stats)
from freqtrade.edge import PairInfo
from freqtrade.enums import ExitType
from freqtrade.optimize.optimize_reports import (_get_resample_from_period, generate_backtest_stats,
generate_daily_stats, generate_edge_table,
generate_exit_reason_stats, generate_pair_metrics,
generate_periodic_breakdown_stats,
generate_strategy_comparison,
generate_trading_stats, show_sorted_pairlist,
store_backtest_stats, text_table_bt_results,
text_table_exit_reason, text_table_strategy)
from freqtrade.resolvers.strategy_resolver import StrategyResolver
from tests.conftest import CURRENT_TEST_STRATEGY
from tests.data.test_history import _backup_file, _clean_test_file
def test_text_table_bt_results():
results = pd.DataFrame(
{
'pair': ['ETH/BTC', 'ETH/BTC', 'ETH/BTC'],
'profit_ratio': [0.1, 0.2, -0.05],
'profit_abs': [0.2, 0.4, -0.1],
'trade_duration': [10, 30, 20],
}
)
result_str = (
'| Pair | Buys | Avg Profit % | Cum Profit % | Tot Profit BTC | Tot Profit % |'
' Avg Duration | Win Draw Loss Win% |\n'
'|---------+--------+----------------+----------------+------------------+----------------+'
'----------------+-------------------------|\n'
'| ETH/BTC | 3 | 8.33 | 25.00 | 0.50000000 | 12.50 |'
' 0:20:00 | 2 0 1 66.7 |\n'
'| TOTAL | 3 | 8.33 | 25.00 | 0.50000000 | 12.50 |'
' 0:20:00 | 2 0 1 66.7 |'
)
pair_results = generate_pair_metrics(['ETH/BTC'], stake_currency='BTC',
starting_balance=4, results=results)
assert text_table_bt_results(pair_results, stake_currency='BTC') == result_str
def test_generate_backtest_stats(default_conf, testdatadir, tmpdir):
default_conf.update({'strategy': CURRENT_TEST_STRATEGY})
StrategyResolver.load_strategy(default_conf)
results = {'DefStrat': {
'results': pd.DataFrame({"pair": ["UNITTEST/BTC", "UNITTEST/BTC",
"UNITTEST/BTC", "UNITTEST/BTC"],
"profit_ratio": [0.003312, 0.010801, 0.013803, 0.002780],
"profit_abs": [0.000003, 0.000011, 0.000014, 0.000003],
"open_date": [Arrow(2017, 11, 14, 19, 32, 00).datetime,
Arrow(2017, 11, 14, 21, 36, 00).datetime,
Arrow(2017, 11, 14, 22, 12, 00).datetime,
Arrow(2017, 11, 14, 22, 44, 00).datetime],
"close_date": [Arrow(2017, 11, 14, 21, 35, 00).datetime,
Arrow(2017, 11, 14, 22, 10, 00).datetime,
Arrow(2017, 11, 14, 22, 43, 00).datetime,
Arrow(2017, 11, 14, 22, 58, 00).datetime],
"open_rate": [0.002543, 0.003003, 0.003089, 0.003214],
"close_rate": [0.002546, 0.003014, 0.003103, 0.003217],
"trade_duration": [123, 34, 31, 14],
"is_open": [False, False, False, True],
"is_short": [False, False, False, False],
"stake_amount": [0.01, 0.01, 0.01, 0.01],
"exit_reason": [ExitType.ROI, ExitType.STOP_LOSS,
ExitType.ROI, ExitType.FORCE_EXIT]
}),
'config': default_conf,
'locks': [],
'final_balance': 1000.02,
'rejected_signals': 20,
'timedout_entry_orders': 0,
'timedout_exit_orders': 0,
'backtest_start_time': Arrow.utcnow().int_timestamp,
'backtest_end_time': Arrow.utcnow().int_timestamp,
'run_id': '123',
}
}
timerange = TimeRange.parse_timerange('1510688220-1510700340')
min_date = Arrow.fromtimestamp(1510688220)
max_date = Arrow.fromtimestamp(1510700340)
btdata = history.load_data(testdatadir, '1m', ['UNITTEST/BTC'], timerange=timerange,
fill_up_missing=True)
stats = generate_backtest_stats(btdata, results, min_date, max_date)
assert isinstance(stats, dict)
assert 'strategy' in stats
assert 'DefStrat' in stats['strategy']
assert 'strategy_comparison' in stats
strat_stats = stats['strategy']['DefStrat']
assert strat_stats['backtest_start'] == min_date.strftime(DATETIME_PRINT_FORMAT)
assert strat_stats['backtest_end'] == max_date.strftime(DATETIME_PRINT_FORMAT)
assert strat_stats['total_trades'] == len(results['DefStrat']['results'])
# Above sample had no loosing trade
assert strat_stats['max_drawdown_account'] == 0.0
# Retry with losing trade
results = {'DefStrat': {
'results': pd.DataFrame(
{"pair": ["UNITTEST/BTC", "UNITTEST/BTC", "UNITTEST/BTC", "UNITTEST/BTC"],
"profit_ratio": [0.003312, 0.010801, -0.013803, 0.002780],
"profit_abs": [0.000003, 0.000011, -0.000014, 0.000003],
"open_date": [Arrow(2017, 11, 14, 19, 32, 00).datetime,
Arrow(2017, 11, 14, 21, 36, 00).datetime,
Arrow(2017, 11, 14, 22, 12, 00).datetime,
Arrow(2017, 11, 14, 22, 44, 00).datetime],
"close_date": [Arrow(2017, 11, 14, 21, 35, 00).datetime,
Arrow(2017, 11, 14, 22, 10, 00).datetime,
Arrow(2017, 11, 14, 22, 43, 00).datetime,
Arrow(2017, 11, 14, 22, 58, 00).datetime],
"open_rate": [0.002543, 0.003003, 0.003089, 0.003214],
"close_rate": [0.002546, 0.003014, 0.0032903, 0.003217],
"trade_duration": [123, 34, 31, 14],
"is_open": [False, False, False, True],
"is_short": [False, False, False, False],
"stake_amount": [0.01, 0.01, 0.01, 0.01],
"exit_reason": [ExitType.ROI, ExitType.ROI,
ExitType.STOP_LOSS, ExitType.FORCE_EXIT]
}),
'config': default_conf,
'locks': [],
'final_balance': 1000.02,
'rejected_signals': 20,
'timedout_entry_orders': 0,
'timedout_exit_orders': 0,
'backtest_start_time': Arrow.utcnow().int_timestamp,
'backtest_end_time': Arrow.utcnow().int_timestamp,
'run_id': '124',
}
}
stats = generate_backtest_stats(btdata, results, min_date, max_date)
assert isinstance(stats, dict)
assert 'strategy' in stats
assert 'DefStrat' in stats['strategy']
assert 'strategy_comparison' in stats
strat_stats = stats['strategy']['DefStrat']
assert pytest.approx(strat_stats['max_drawdown_account']) == 1.399999e-08
assert strat_stats['drawdown_start'] == '2017-11-14 22:10:00'
assert strat_stats['drawdown_end'] == '2017-11-14 22:43:00'
assert strat_stats['drawdown_end_ts'] == 1510699380000
assert strat_stats['drawdown_start_ts'] == 1510697400000
assert strat_stats['pairlist'] == ['UNITTEST/BTC']
# Test storing stats
filename = Path(tmpdir / 'btresult.json')
filename_last = Path(tmpdir / LAST_BT_RESULT_FN)
_backup_file(filename_last, copy_file=True)
assert not filename.is_file()
store_backtest_stats(filename, stats)
# get real Filename (it's btresult-<date>.json)
last_fn = get_latest_backtest_filename(filename_last.parent)
assert re.match(r"btresult-.*\.json", last_fn)
filename1 = Path(tmpdir / last_fn)
assert filename1.is_file()
content = filename1.read_text()
assert 'max_drawdown_account' in content
assert 'strategy' in content
assert 'pairlist' in content
assert filename_last.is_file()
_clean_test_file(filename_last)
filename1.unlink()
def test_store_backtest_stats(testdatadir, mocker):
dump_mock = mocker.patch('freqtrade.optimize.optimize_reports.file_dump_json')
store_backtest_stats(testdatadir, {'metadata': {}})
assert dump_mock.call_count == 3
assert isinstance(dump_mock.call_args_list[0][0][0], Path)
assert str(dump_mock.call_args_list[0][0][0]).startswith(str(testdatadir / 'backtest-result'))
dump_mock.reset_mock()
filename = testdatadir / 'testresult.json'
store_backtest_stats(filename, {'metadata': {}})
assert dump_mock.call_count == 3
assert isinstance(dump_mock.call_args_list[0][0][0], Path)
# result will be testdatadir / testresult-<timestamp>.json
assert str(dump_mock.call_args_list[0][0][0]).startswith(str(testdatadir / 'testresult'))
def test_generate_pair_metrics():
results = pd.DataFrame(
{
'pair': ['ETH/BTC', 'ETH/BTC'],
'profit_ratio': [0.1, 0.2],
'profit_abs': [0.2, 0.4],
'trade_duration': [10, 30],
'wins': [2, 0],
'draws': [0, 0],
'losses': [0, 0]
}
)
pair_results = generate_pair_metrics(['ETH/BTC'], stake_currency='BTC',
starting_balance=2, results=results)
assert isinstance(pair_results, list)
assert len(pair_results) == 2
assert pair_results[-1]['key'] == 'TOTAL'
assert (
pytest.approx(pair_results[-1]['profit_mean_pct']) == pair_results[-1]['profit_mean'] * 100)
assert (
pytest.approx(pair_results[-1]['profit_sum_pct']) == pair_results[-1]['profit_sum'] * 100)
def test_generate_daily_stats(testdatadir):
filename = testdatadir / "backtest-result_new.json"
bt_data = load_backtest_data(filename)
res = generate_daily_stats(bt_data)
assert isinstance(res, dict)
assert round(res['backtest_best_day'], 4) == 0.1796
assert round(res['backtest_worst_day'], 4) == -0.1468
assert res['winning_days'] == 19
assert res['draw_days'] == 0
assert res['losing_days'] == 2
# Select empty dataframe!
res = generate_daily_stats(bt_data.loc[bt_data['open_date'] == '2000-01-01', :])
assert isinstance(res, dict)
assert round(res['backtest_best_day'], 4) == 0.0
assert res['winning_days'] == 0
assert res['draw_days'] == 0
assert res['losing_days'] == 0
def test_generate_trading_stats(testdatadir):
filename = testdatadir / "backtest-result_new.json"
bt_data = load_backtest_data(filename)
res = generate_trading_stats(bt_data)
assert isinstance(res, dict)
assert res['winner_holding_avg'] == timedelta(seconds=1440)
assert res['loser_holding_avg'] == timedelta(days=1, seconds=21420)
assert 'wins' in res
assert 'losses' in res
assert 'draws' in res
# Select empty dataframe!
res = generate_trading_stats(bt_data.loc[bt_data['open_date'] == '2000-01-01', :])
assert res['wins'] == 0
assert res['losses'] == 0
def test_text_table_exit_reason():
results = pd.DataFrame(
{
'pair': ['ETH/BTC', 'ETH/BTC', 'ETH/BTC'],
'profit_ratio': [0.1, 0.2, -0.1],
'profit_abs': [0.2, 0.4, -0.2],
'trade_duration': [10, 30, 10],
'wins': [2, 0, 0],
'draws': [0, 0, 0],
'losses': [0, 0, 1],
'exit_reason': [ExitType.ROI, ExitType.ROI, ExitType.STOP_LOSS]
}
)
result_str = (
'| Exit Reason | Exits | Win Draws Loss Win% | Avg Profit % | Cum Profit % |'
' Tot Profit BTC | Tot Profit % |\n'
'|---------------+---------+--------------------------+----------------+----------------+'
'------------------+----------------|\n'
'| roi | 2 | 2 0 0 100 | 15 | 30 |'
' 0.6 | 15 |\n'
'| stop_loss | 1 | 0 0 1 0 | -10 | -10 |'
' -0.2 | -5 |'
)
exit_reason_stats = generate_exit_reason_stats(max_open_trades=2,
results=results)
assert text_table_exit_reason(exit_reason_stats=exit_reason_stats,
stake_currency='BTC') == result_str
def test_generate_sell_reason_stats():
results = pd.DataFrame(
{
'pair': ['ETH/BTC', 'ETH/BTC', 'ETH/BTC'],
'profit_ratio': [0.1, 0.2, -0.1],
'profit_abs': [0.2, 0.4, -0.2],
'trade_duration': [10, 30, 10],
'wins': [2, 0, 0],
'draws': [0, 0, 0],
'losses': [0, 0, 1],
'exit_reason': [ExitType.ROI.value, ExitType.ROI.value, ExitType.STOP_LOSS.value]
}
)
exit_reason_stats = generate_exit_reason_stats(max_open_trades=2,
results=results)
roi_result = exit_reason_stats[0]
assert roi_result['exit_reason'] == 'roi'
assert roi_result['trades'] == 2
assert pytest.approx(roi_result['profit_mean']) == 0.15
assert roi_result['profit_mean_pct'] == round(roi_result['profit_mean'] * 100, 2)
assert pytest.approx(roi_result['profit_mean']) == 0.15
assert roi_result['profit_mean_pct'] == round(roi_result['profit_mean'] * 100, 2)
stop_result = exit_reason_stats[1]
assert stop_result['exit_reason'] == 'stop_loss'
assert stop_result['trades'] == 1
assert pytest.approx(stop_result['profit_mean']) == -0.1
assert stop_result['profit_mean_pct'] == round(stop_result['profit_mean'] * 100, 2)
assert pytest.approx(stop_result['profit_mean']) == -0.1
assert stop_result['profit_mean_pct'] == round(stop_result['profit_mean'] * 100, 2)
def test_text_table_strategy(testdatadir):
filename = testdatadir / "backtest-result_multistrat.json"
bt_res_data = load_backtest_stats(filename)
bt_res_data_comparison = bt_res_data.pop('strategy_comparison')
result_str = (
'| Strategy | Buys | Avg Profit % | Cum Profit % | Tot Profit BTC |'
' Tot Profit % | Avg Duration | Win Draw Loss Win% | Drawdown |\n'
'|----------------+--------+----------------+----------------+------------------+'
'----------------+----------------+-------------------------+-----------------------|\n'
'| StrategyTestV2 | 179 | 0.08 | 14.39 | 0.02608550 |'
' 260.85 | 3:40:00 | 170 0 9 95.0 | 0.00308222 BTC 8.67% |\n'
'| TestStrategy | 179 | 0.08 | 14.39 | 0.02608550 |'
' 260.85 | 3:40:00 | 170 0 9 95.0 | 0.00308222 BTC 8.67% |'
)
strategy_results = generate_strategy_comparison(bt_stats=bt_res_data['strategy'])
assert strategy_results == bt_res_data_comparison
assert text_table_strategy(strategy_results, 'BTC') == result_str
def test_generate_edge_table():
results = {}
results['ETH/BTC'] = PairInfo(-0.01, 0.60, 2, 1, 3, 10, 60)
assert generate_edge_table(results).count('+') == 7
assert generate_edge_table(results).count('| ETH/BTC |') == 1
assert generate_edge_table(results).count(
'| Risk Reward Ratio | Required Risk Reward | Expectancy |') == 1
def test_generate_periodic_breakdown_stats(testdatadir):
filename = testdatadir / "backtest-result_new.json"
bt_data = load_backtest_data(filename).to_dict(orient='records')
res = generate_periodic_breakdown_stats(bt_data, 'day')
assert isinstance(res, list)
assert len(res) == 21
day = res[0]
assert 'date' in day
assert 'draws' in day
assert 'loses' in day
assert 'wins' in day
assert 'profit_abs' in day
# Select empty dataframe!
res = generate_periodic_breakdown_stats([], 'day')
assert res == []
def test__get_resample_from_period():
assert _get_resample_from_period('day') == '1d'
assert _get_resample_from_period('week') == '1w'
assert _get_resample_from_period('month') == '1M'
with pytest.raises(ValueError, match=r"Period noooo is not supported."):
_get_resample_from_period('noooo')
def test_show_sorted_pairlist(testdatadir, default_conf, capsys):
filename = testdatadir / "backtest-result_new.json"
bt_data = load_backtest_stats(filename)
default_conf['backtest_show_pair_list'] = True
show_sorted_pairlist(default_conf, bt_data)
out, err = capsys.readouterr()
assert 'Pairs for Strategy StrategyTestV3: \n[' in out
assert 'TOTAL' not in out
assert '"ETH/BTC", // ' in out