def test_single_number():
order_book_ids = ["000001.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids, start_date="2019-01-01", end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data, sequence_window=sequence_window, add_cash=True)
state = env.reset()
h_t_list = []
orderlist = [0.5, 0.8, 0.4, 1.0, 0.0]
for i in range(len(orderlist)):
next_state, reward, done, info = env.step(orderlist[i])
h_t_list.append(info["h_t"])
'''
stock_reward = [0.0219, -0.0103, 0.0175, -0.0017, -0.0039, 0.0059]
v_t = portfolio_value = [1000000, 994900.0, 1008848.5, 1008223.0, 1004290.9, 1004391.4]
expected_portfolio calculate: portfolio_value[i]*weight*mock_data["returns"][i+1] for stock and cash
000001.XSHE 2019-01-01 0.0219
2019-01-02 -0.0103
2019-01-03 0.0175
2019-01-04 -0.0017
2019-01-05 -0.0039
2019-01-06 0.0059
2019-01-07 -0.0049
2019-01-08 -0.0003
2019-01-09 -0.0136
2019-01-10 0.0068
2019-01-11 0.0077
2019-01-12 0.0136
2019-01-13 -0.0022
2019-01-14 -0.0012
'''
expected_portfolio = ([494850, 500050], [809848.6, 198999.898], [402853.3822, 605369.6309], [1004290.9, 0], [0, 1004391.359])
np.testing.assert_almost_equal(h_t_list, expected_portfolio, decimal=1)
def test_single_number_R():
order_book_ids = ["000001.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids,
start_date="2019-01-01",
end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data,
sequence_window=sequence_window,
add_cash=True)
state = env.reset()
h_t_list = []
orderlist = [0.5, 0.8, 0.4, 1.0, 0.0, -0.5, -0.7, -0.4, -1.0, 0.3]
for i in range(len(orderlist)):
next_state, reward, done, info = env.step(orderlist[i])
h_t_list.append(info["h_t"])
expected_h_t = ([494850,
500050], [809848.6,
198999.898], [402853.3822,
605369.6309], [1004290.9, 0],
[0,
1004391.359], [-499734.9212,
1506737.699], [-704690.4898, 1712075.95],
[-397473.9834,
1410480.594], [-1019895.146,
2026216.001], [304220.9, 704495.1425])
np.testing.assert_almost_equal(h_t_list, expected_h_t, decimal=1)
class TestDataWraper(unittest.TestCase):
def setUp(self):
self.order_book_id_number = 100
self.feature_number = 10
self.toy_data = create_toy_data(order_book_ids_number=order_book_id_number, feature_number=self.feature_number, start="2019-05-01", end="2019-12-12", frequency="D")
self.env_2d = PortfolioTradingGym(data_df=self.toy_data, sequence_window=1, add_cash=False, mode="numpy")
self.sequence_window =3
self.env_3d = PortfolioTradingGym(data_df = self.toy_data, sequence_window=self.sequence_window, add_cash=False, mode="numpy")
def test_sequence_window_equal_1(self):
state = self.env_2d.reset()
self.assertEqual(state.shape,(self.order_book_id_number, self.feature_number))
random_action = self.env_2d.action_space.sample()
next_state, reward, done, info = self.env_2d.step(random_action)
self.assertEqual(next_state.shape, (self.order_book_id_number, self.feature_number))
return next_state
def test_sequence_window_mt_1(self):
state = self.env_3d.reset()
self.assertEqual(state.shape,(self.order_book_id_number, self.sequence_window, self.feature_number))
random_action = self.env_3d.action_space.sample()
next_state, reward, done, info = self.env_3d.step(random_action)
self.assertEqual(next_state.shape, (self.order_book_id_number, self.sequence_window, self.feature_number))
def test_torch_wrapper(self):
def setUp(self):
self.order_book_id_number = 100
self.feature_number = 10
self.toy_data = create_toy_data(order_book_ids_number=order_book_id_number, feature_number=self.feature_number, start="2019-05-01", end="2019-12-12", frequency="D")
self.env_2d = PortfolioTradingGym(data_df=self.toy_data, sequence_window=1, add_cash=False, mode="numpy")
self.sequence_window =3
self.env_3d = PortfolioTradingGym(data_df = self.toy_data, sequence_window=self.sequence_window, add_cash=False, mode="numpy")
def test_positive():
order_book_ids = ["000001.XSHE", "000002.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids,
start_date="2019-01-01",
end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data,
sequence_window=sequence_window,
add_cash=True)
state = env.reset()
h_t_list = []
orderlista = [0.5, 0.6, 0.2, 1.0, 0.0, 0.0, 0.4, 0.1]
orderlistb = [0.5, 0.1, 0.7, 0.0, 1.0, 0.0, 0.6, 0.2]
for i in range(len(orderlista)):
next_state, reward, done, info = env.step(
[orderlista[i], orderlistb[i], 0])
h_t_list.append(info["h_t"])
'''
000001.XSHE 2019-01-01 0.0219
2019-01-02 -0.0103
2019-01-03 0.0175
2019-01-04 -0.0017
2019-01-05 -0.0039
2019-01-06 0.0059
2019-01-07 -0.0049
2019-01-08 -0.0003
2019-01-09 -0.0136
000002.XSHE 2019-01-01 -0.0186
2019-01-02 0.0098
2019-01-03 -0.0030
2019-01-04 0.0065
2019-01-05 0.0111
2019-01-06 0.0006
2019-01-07 0.0112
2019-01-08 -0.0304
2019-01-09 -0.0094
'''
expected_h_t = ([494850, 504900,
0.0], [610347.375, 99675.075, 299954.9925
], [201652.0962, 711579.5973,
101007.84], [1010283.99, 0.0,
0.0], [0.0, 1010890.17, 0.0],
[0.0, 0.0,
1010991.259], [404275.1847, 588154.2748,
0.0], [97893.241, 196620.0124, 694770.1])
np.testing.assert_almost_equal(h_t_list, expected_h_t, decimal=1)
def test_single_array():
order_book_ids = ["000001.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids, start_date="2019-01-01", end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data, sequence_window=sequence_window, add_cash=True)
state = env.reset()
h_t_list = []
orderlist = [0.5, 0.8, 0.4, 1.0, 0.0]
for i in range(len(orderlist)):
next_state, reward, done, info = env.step((orderlist[i], 0))
h_t_list.append(info["h_t"])
'''
Action for "cash" is never used. Whenever add_cash = true, next "cash" is calculated by negative sum of the other
trade (in market_simulator.py). Then action for cash can be any number and the results are still right.
'''
expected_portfolio = ([494850, 500050], [809848.6, 198999.898], [402853.3822, 605369.6309], [1004290.9, 0], [0, 1004391.359])
np.testing.assert_almost_equal(h_t_list, expected_portfolio, decimal=1)
def test_single_cash_false():
order_book_ids = ["000001.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids, start_date="2019-01-01", end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data, sequence_window=sequence_window, add_cash=False)
state = env.reset()
h_t_list = []
orderlist = [0.5, 0.8, 0.4, 1.0, 0.0]
for i in range(len(orderlist)):
next_state, reward, done, info = env.step(orderlist[i])
h_t_list.append(info["h_t"])
'''
Portfolio value becomes stock value incorrectly after every step.
When portfolio value = 0, self.w_t (h_next / sum(h_next) in portfolio_gym) is NaN, then trigger value error in
market simulator.
'''
expected_portfolio = ([494850], [809848.6], [402853.3822], [1004290.9], [0])
np.testing.assert_almost_equal(h_t_list, expected_portfolio, decimal=1)
def test_single_cashfalse_R():
order_book_ids = ["000001.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids,
start_date="2019-01-01",
end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data,
sequence_window=sequence_window,
add_cash=False)
state = env.reset()
h_t_list = []
orderlist = [0.5, 0.8, 0.4, 1.0, 0.0, -0.5, -0.7, -0.4, -1.0, 0.3]
for i in range(len(orderlist)):
next_state, reward, done, info = env.step(orderlist[i])
h_t_list.append(info["h_t"])
expected_h_t = ([494850], [809848.6], [402853.3822
], [1004290.9], [0], [-499734.9212],
[-704690.4898], [-397473.9834], [-1019895.146], [304220.9])
np.testing.assert_almost_equal(h_t_list, expected_h_t, decimal=1)
def test_mul_R():
order_book_ids = ["000001.XSHE", "000002.XSHE"]
mock_data = create_mock_data(order_book_ids=order_book_ids,
start_date="2019-01-01",
end_date="2019-01-14")
sequence_window = 1
env = PortfolioTradingGym(data_df=mock_data,
sequence_window=sequence_window,
add_cash=True)
state = env.reset()
h_t_list = []
orderlista = [0.5, 0.6, -0.2, -0.3, -1.0, 1.0]
orderlistb = [0.5, -0.2, 0.7, -0.4, 1.0, -1.0]
for i in range(len(orderlista)):
next_state, reward, done, info = env.step(
[orderlista[i], orderlistb[i], 0])
h_t_list.append(info["h_t"])
'''
000001.XSHE 2019-01-01 0.0219
2019-01-02 -0.0103
2019-01-03 0.0175
2019-01-04 -0.0017
2019-01-05 -0.0039
2019-01-06 0.0059
2019-01-07 -0.0049
000002.XSHE 2019-01-01 -0.0186
2019-01-02 0.0098
2019-01-03 -0.0030
2019-01-04 0.0065
2019-01-05 0.0111
2019-01-06 0.0006
2019-01-07 0.0112
'''
expected_h_t = ([494850, 504900,
0.0], [610347.375, -199350.15, 599909.985],
[-201837.7335, 712234.6748,
505504.15], [-303581.7231, -410871.0374, 1727204.558],
[-1018727.033, 1013359.449,
1012853.07], [1002548.843, -1018769.3595, 1007586.27])
print(h_t_list)
np.testing.assert_almost_equal(h_t_list, expected_h_t, decimal=1)
def test_invest_on_first_stock():
order_book_ids = ["000001.XSHE","600000.XSHG"]
mock_data = create_mock_data(order_book_ids=order_book_ids, start_date="2019-01-01", end_date="2019-06-11")
sequence_window=3
env = PortfolioTradingGym(data_df=mock_data, sequence_window=sequence_window, add_cash=True)
state = env.reset()
# the first stock buy and hold strategy
action = np.array([1.0, 0.0, 0.0])
while True:
next_state, reward, done, info = env.step(action)
if done:
break
portfolio_reward = np.array(env.experience_buffer["reward"])
expected_reward = mock_data.xs("000001.XSHE")["returns"].iloc[sequence_window:].values
#pdb.set_trace()
np.testing.assert_almost_equal(portfolio_reward, expected_reward, decimal=3)
#env.render()
return mock_data, env
def test_equal_weighted_portfolio():
order_book_ids = ["000001.XSHE","600000.XSHG"]
mock_data = create_mock_data(order_book_ids=order_book_ids, start_date="2019-01-01", end_date="2019-06-11")
sequence_window=3
env = PortfolioTradingGym(data_df=mock_data, sequence_window=sequence_window, add_cash=True)
state = env.reset()
print(state)
action = np.array([0.5, 0.5, 0])
while True:
next_state, reward, done, info = env.step(action)
if done:
break
portfolio_reward = np.array(env.experience_buffer["reward"])
expected_reward = np.array(env.experience_buffer["reward_benchmark"])
np.testing.assert_almost_equal(portfolio_reward, expected_reward, decimal=3)
env.render()
return mock_data, env
# -*- coding: utf-8 -*-
import pdb
import numpy as np
import pandas as pd
from sklearn.linear_model import LinearRegression
from trading_gym.utils.data.toy import create_toy_data
from trading_gym.envs.portfolio_gym.portfolio_gym import PortfolioTradingGym
order_book_id_number = 100
toy_data = create_toy_data(order_book_ids_number=order_book_id_number,
feature_number=10,
start="2019-05-01",
end="2019-12-12",
frequency="D")
env = PortfolioTradingGym(data_df=toy_data, sequence_window=1, add_cash=False)
state = env.reset()
while True:
next_state, reward, done, info = env.step(action=None)
label = info["one_step_fwd_returns"]
print(state)
print(label)
#
regressor = LinearRegression()
regressor.fit(state.values, label.values)
#display and store
print(regressor.coef_)
env.experience_buffer["coef"].append(regressor.coef_)
frequency = "1d"
#
data = history_bars(order_book_ids=order_book_ids,
bar_count=bar_count + rolling_window,
frequency=frequency,
fields=fields,
dt=dt)
pct_change = data.groupby(level="order_book_id")["close"].pct_change()
pct_change.name = "returns"
feature_df = pct_change.to_frame()
feature_df = feature_df.fillna(0)
# the first "returns" as state, the last column is feed into environment serving as to calculate next_fwd_returns
feature_df["label"] = feature_df["returns"]
env = PortfolioTradingGym(data_df=feature_df,
sequence_window=20,
add_cash=False)
# ============================================================================#
# step2: create a policy, policy is a funtion mapping states to action #
# =========================================================================== #
def policy(state: pd.DataFrame) -> pd.DataFrame:
volatility = state.groupby(level=0).std()
volatility_reverse = 1 / volatility
weight = volatility_reverse / volatility_reverse.sum()
weight.name = "weight"
return weight
# =================================================== #
import numpy as np import pandas as pd from sklearn.linear_model import LinearRegression from trading_gym.utils.data.toy import create_toy_data from trading_gym.envs.portfolio_gym.portfolio_gym import PortfolioTradingGym order_book_id_number = 100 toy_data = create_toy_data(order_book_ids_number=order_book_id_number, feature_number=10, start="2019-05-01", end="2019-12-12", frequency="D") env1 = PortfolioTradingGym(data_df=toy_data, sequence_window=1, add_cash=False, mode="pandas") state1 = env1.reset() env2 = PortfolioTradingGym(data_df=toy_data, sequence_window=3, add_cash=False, mode="pandas") state2 = env2.reset()