def test_difference_not_inplace(self, data_frame, exchange):
transformer = FractionalDifference(difference_order=0.5, inplace=False)
transformed_frame = transformer.transform(data_frame,
exchange.generated_space)
expected_data_frame = pd.DataFrame([{
'open': 100,
'low': 50,
'high': 250,
'close': 200,
'open_diff_0.5': -26.20469322,
'low_diff_0.5': -46.15180724,
'high_diff_0.5': 33.63664884,
'close_diff_0.5': 13.68953482,
}, {
'open': 200,
'low': 150,
'high': 350,
'close': 300,
'open_diff_0.5': 134.53651465,
'low_diff_0.5': 118.24976426,
'high_diff_0.5': 183.39676584,
'close_diff_0.5': 167.11001545,
}])
assert np.allclose(expected_data_frame.values,
transformed_frame.values)
def test_incremental_difference_inplace_false(self, data_frame,
reference_frame):
frame_tail = data_frame.tail(200)
transformer = FractionalDifference(difference_order=0.5, inplace=False)
transformed_frame = transformer.transform(frame_tail)
assert transformed_frame is not None
assert isinstance(frame_tail, pd.DataFrame)
def test_transform_space(self, data_frame):
transformer = FractionalDifference(
difference_order=0.5, inplace=False, all_column_names=data_frame.columns)
low = np.array([1E-3, ] * 4 + [1E-3, ])
high = np.array([1E3, ] * 4 + [1E3, ])
input_space = Box(low=low, high=high, dtype=np.float16)
transformed_space = transformer.transform_space(input_space)
assert transformed_space != input_space
def test_incremental_difference(self, data_frame, exchange):
transformer = FractionalDifference(difference_order=0.5, inplace=True)
transformed_frame = transformer.transform(data_frame,
exchange.generated_space)
expected_data_frame = pd.DataFrame([{
'open': -26.20469322,
'low': -46.15180724,
'high': 33.63664884,
'close': 13.68953482,
}, {
'open': 134.53651465,
'low': 118.24976426,
'high': 183.39676584,
'close': 167.11001545,
}])
assert np.allclose(expected_data_frame.values,
transformed_frame.values)
next_frame = pd.DataFrame([{
'open': 200,
'low': 150,
'high': 350,
'close': 300,
}, {
'open': 300,
'low': 250,
'high': 450,
'close': 400,
}])
transformed_frame = transformer.transform(next_frame,
exchange.generated_space)
expected_data_frame = pd.DataFrame([{
'open': 127.785105,
'low': 87.031409,
'high': 250.046192,
'close': 209.292496,
}, {
'open': 185.484853,
'low': 166.817514,
'high': 241.486873,
'close': 222.819533,
}])
assert np.allclose(expected_data_frame.values,
transformed_frame.values)
def test_incremental_difference(self, data_frame, reference_frame):
frame_tail = data_frame.tail(200)
transformer = FractionalDifference(difference_order=0.5, inplace=True)
significant = 3
transformed_frame = transformer.transform(frame_tail)
reference_frame = reference_frame.round(significant)
transformed_frame = transformed_frame.round(significant)
close_1 = transformed_frame.Close.values
close_2 = reference_frame.Close.values
is_valid = (close_1 == close_2).all()
assert transformed_frame is not None
assert is_valid
def test_select_correct_columns(self, data_frame, exchange):
transformer = FractionalDifference(
columns=['open', 'close'], difference_order=0.5, inplace=True)
transformed_frame = transformer.transform(data_frame, exchange.generated_space)
expected_data_frame = pd.DataFrame([{
'open': -26.20469322,
'low': 50,
'high': 250,
'close': 13.68953482,
},
{
'open': 134.53651465,
'low': 150,
'high': 350,
'close': 167.11001545,
}])
assert np.allclose(expected_data_frame.values, transformed_frame.values)
def test_difference_inplace(self, data_frame):
transformer = FractionalDifference(
difference_order=0.5, inplace=True, all_column_names=data_frame.columns)
transformed_frame = transformer.transform(data_frame)
expected_data_frame = pd.DataFrame([{
'open': -26.20469322,
'low': -46.15180724,
'high': 33.63664884,
'close': 13.68953482,
},
{
'open': 134.53651465,
'low': 118.24976426,
'high': 183.39676584,
'close': 167.11001545,
}])
assert np.allclose(expected_data_frame.values, transformed_frame.values)
def test_incremental_transform(self, data_frame):
difference_all = FractionalDifference(
difference_order=0.5,
inplace=True,
all_column_names=data_frame.columns)
feature_pipeline = FeaturePipeline(steps=[difference_all])
transformed_frame = feature_pipeline.transform(data_frame)
expected_data_frame = pd.DataFrame([{
'open': -26.20469322,
'low': -46.15180724,
'high': 33.63664884,
'close': 13.68953482,
}, {
'open': 134.53651465,
'low': 118.24976426,
'high': 183.39676584,
'close': 167.11001545,
}])
assert np.allclose(expected_data_frame.values,
transformed_frame.values)
next_frame = pd.DataFrame([{
'open': 200,
'low': 150,
'high': 350,
'close': 300,
}, {
'open': 300,
'low': 250,
'high': 450,
'close': 400,
}])
transformed_frame = feature_pipeline.transform(next_frame)
expected_data_frame = pd.DataFrame([{
'open': 127.785105,
'low': 87.031409,
'high': 250.046192,
'close': 209.292496,
}, {
'open': 185.484853,
'low': 166.817514,
'high': 241.486873,
'close': 222.819533,
}])
assert np.allclose(expected_data_frame.values,
transformed_frame.values)
def test_transform_space(self, data_frame, exchange):
difference_all = FractionalDifference(difference_order=0.5, inplace=False)
feature_pipeline = FeaturePipeline(steps=[difference_all])
low = np.array([1E-3, ] * 4 + [1E-3, ])
high = np.array([1E3, ] * 4 + [1E3, ])
input_space = Box(low=low, high=high, dtype=np.float16)
transformed_space = feature_pipeline.transform_space(
input_space, exchange.generated_columns)
assert transformed_space != input_space
def get_env(file_path):
df = load_data(file_path)
normalize = MinMaxNormalizer(inplace=True)
difference = FractionalDifference(difference_order=0.6, inplace=True)
feature_pipeline = FeaturePipeline(steps=[normalize, difference])
reward_strategy = SimpleProfitStrategy()
action_strategy = DiscreteActionStrategy(n_actions=20, instrument_symbol='BTC/USDT')
exchange = SimulatedExchange(base_instrument='USDT',
should_pretransform_obs=True,
feature_pipeline=feature_pipeline
)
exchange.data_frame = df[:STEP]
environment = TradingEnvironment(exchange=exchange,
action_strategy=action_strategy,
reward_strategy=reward_strategy,
feature_pipeline=feature_pipeline)
return environment
import tensorflow
assert tensorflow.__version__ == '1.13.1'
from tensortrade.strategies import TensorforceTradingStrategy
from tensortrade.rewards import DirectProfitStrategy
from tensortrade.actions import FutureActionStrategy
reward_strategy = DirectProfitStrategy()
action_strategy = FutureActionStrategy()
#%% Feature Pipeline
from tensortrade.features.stationarity import FractionalDifference
from tensortrade.features.scalers import MinMaxNormalizer
from tensortrade.features import FeaturePipeline
normalize = MinMaxNormalizer(inplace=True)
difference = FractionalDifference(difference_order=0.6, inplace=True)
feature_pipeline = FeaturePipeline(steps=[])
#%% Data Input
from tensortrade.exchanges.simulated.future_exchange import FutureExchange
data = pd.read_csv('TA.csv', index_col=0)
data = data[data.index % 30 == 0]
data = data.reset_index(drop=True)
exchange = FutureExchange(data,
base_instrument='RMB',
exclude_close=True,
initial_balance=10000,
should_pretransform_obs=False)
df = pd.read_csv('./data/btc_ohclv_1h.csv')
exchange = SimulatedExchange(data_frame=df, base_instrument='USD')
# Feature Pipelines
from tensortrade.features import FeaturePipeline
from tensortrade.features.scalers import MinMaxNormalizer
from tensortrade.features.stationarity import FractionalDifference
from tensortrade.features.indicators import SimpleMovingAverage
price_columns = ['open', 'high', 'low', 'close']
normalize_price = MinMaxNormalizer(price_columns)
moving_averages = SimpleMovingAverage(price_columns)
difference_all = FractionalDifference(difference_order=0.6)
feature_pipeline = FeaturePipeline(
steps=[normalize_price, moving_averages, difference_all])
exchange.feature_pipeline = feature_pipeline
# Action Strategies
from tensortrade.actions import DiscreteActionStrategy
action_strategy = DiscreteActionStrategy(n_actions=20, instrument_symbol='BTC')
# Reward Strategies
from tensortrade.rewards import SimpleProfitStrategy