def test_dry_run(self):
""" Test dry run in euler. Check that it tests a model on historical
data properly.
"""
# Initialize model with fixed random state.
model = tree.DecisionTreeRegressor(random_state=888)
# Initialize learner and force load test data.
learner = Learner("GBP_USD")
learner.data_mat = base.read_features(self.tmp_clean_file)
# Build model and test preliminary results.
learner.build_model(model, 0.78)
pred, _ = learner.test_model(model)
# Initialize Euler and force load test data.
strategy = euler.Euler("GBP_USD")
strategy.set_params(unit_shape=common.UNIT_LINEAR, threshold=85)
strategy.test_data = transformer.read_raw_file(self.tmp_raw_file)
balance = strategy.dry_run(pred)
# Test the results.
self.assertEqual(balance.size, 628)
self.assertEqual(round(balance[0], 4), -0.1965)
self.assertEqual(round(balance[210], 4), 0.3467)
self.assertEqual(round(balance[-143], 4), -19.6579)
self.assertEqual(round(balance[-30], 4), -20.7953)
self.assertEqual(round(sum(balance), 4), -4088.2712)
return
def test_leaner_tree_regressor(self):
""" Test the learner builds and tests a tree regressor properly."""
# Initialize model with fixed random state.
model = tree.DecisionTreeRegressor(random_state=888)
# Initialize learner and force load test data.
learner = Learner("GBP_USD")
learner.data_mat = base.read_features(self.tmp_file)
# Build model and test preliminary results.
learner.build_model(model, 0.78)
pred, result = learner.test_model(model)
# Test the results.
self.assertEqual(pred.size, 628)
self.assertEqual(pred[0], -160.2)
self.assertEqual(pred[200], -31.0)
self.assertEqual(pred[-30], -53.6)
self.assertEqual(len(result), 2)
self.assertEqual(round(result['ave_diff'], 4), 88.3980)
self.assertEqual(round(result['prop_op'], 4), 0.4341)
return
def __init__(self, instrument):
""" Initialize the strategy class Euler.
Args:
instrument: string. The currency pair. e.g. 'EUR_USD'.
Returns:
void.
"""
super(Euler, self).__init__(instrument)
# Initialize learner and model.
self.learner = Learner(instrument)
self.model = None
# Read in the raw data file for testing.
test_file = common.get_raw_data(instrument)
self.test_data = transformer.read_raw_file(test_file)
return
class Euler(BaseStrategy):
""" The strategy class Euler. It attempts to predict daily price change
from open to close according to previous day's candle.
"""
# All predictive models used for strategy Euler.
all_models = util.get_all_models()
def __init__(self, instrument):
""" Initialize the strategy class Euler.
Args:
instrument: string. The currency pair. e.g. 'EUR_USD'.
Returns:
void.
"""
super(Euler, self).__init__(instrument)
# Initialize learner and model.
self.learner = Learner(instrument)
self.model = None
# Read in the raw data file for testing.
test_file = common.get_raw_data(instrument)
self.test_data = transformer.read_raw_file(test_file)
return
def set_params(self, **params):
""" Set parameters for strategy Euler.
Args:
params: named parameters for strategy Euler, including:
threshold: positive float. Threshold for the strategy to
take action (either buy or sell).
unit_shape: string. Describes the relationship between the
number of units and the predicted value. One of:
UNIT_CONSTANT, UNIT_LINEAR, UNIT_SQUARE, or UNIT_LOG
as defined in the module common.
Returns:
void.
"""
if 'name' in params:
assert params['name'] == 'Euler'
params['name'] = 'Euler'
self.params = params
return
def parse_controls(self):
""" Determine the SL/TP/TS orders for trade from parameters.
Args:
void.
Returns:
controls: dict. Including take_profit, stop_loss (in price) and
trailing_stop (in pips).
"""
controls = {}
if 'stop_loss' in self.params:
controls['stop_loss'] = self.params['stop_loss']
if 'take_profit' in self.params:
controls['take_profit'] = self.params['take_profit']
if 'trailing_stop' in self.params:
controls['trailing_stop'] = self.params['trailing_stop']
return controls
def parse_units(self, pred):
""" Determine the number of units for trade from parameters.
Args:
pred: float. Predicted change in price.
Returns:
units: signed int. Number of units for trade.
Positive for buy, negative for sell.
"""
# Get the parameters first.
unit_shape = self.params['unit_shape']
threshold = self.params['threshold']
# Take no action if predicted price change under threshold.
if pred > threshold:
sign = 1
elif pred < -threshold:
sign = -1
else:
return 0
# Parse different ways of determining units.
if unit_shape == common.UNIT_CONSTANT:
units = common.CONSTANT_FACTOR
elif unit_shape == common.UNIT_LINEAR:
units = int(abs(pred * common.LINEAR_FACTOR))
elif unit_shape == common.UNIT_SQUARE:
units = int(pred ** 2 * common.SQUARE_FACTOR)
elif unit_shape == common.UNIT_LOG:
units = int(math.log(abs(pred)) * common.LOG_FACTOR)
else:
logger.warn("euler.parse_units(): Unit shape has to be one of \
UNIT_CONSTANT, UNIT_LINEAR, UNIT_SQUARE, or UNIT_LOG \
as defined in 'common'.")
return 0
# Make sure the absoulte value of units doesn't exceed the maximum.
units = min(units, common.MAX_UNITS) * sign
return units
def execute(self, executor, candle):
""" Execute the strategy at day's open.
Args:
executor: exec.Executor. The object for executing trades.
candle: dict. Yesterday's daily candle.
Returns:
void.
"""
# TODO: Report failure.
# Build and format the features first.
features = transformer.candle_to_features(candle, self.pip_factor)
features = np.asarray(features).reshape(1, -1)
# Making and logging the prediction.
pred = self.model.predict(features)
logger.info("Euler: Predicted price change for %s is %.2f.", \
self.instrument, float(pred))
# Parse the parameters.
controls = self.parse_controls()
units = self.parse_units(pred)
# Make the decision.
executor.make_trade(self.instrument, units, **controls)
return
def dry_run(self, pred, **kwargs):
""" Do a dry run of strategy Euler as if the strategy was put in place.
Return the day-to-day balance during the run, and produce a report
on the profitability of this strategy over the given period.
Args:
pred: np.vector. Predicted daily price changes for days in the
last part of the test data.
kwargs: named arguments, including:
print_result: boolean. Whether to print dry run report.
export_plot: string. Name of the plot to be saved.
Returns:
balance: np.array. Accumulated profit/loss of every day.
"""
# Write report title.
report = "\nDry run report: {0}\n\n".format(self.instrument)
report += str(self.params) + '\n'
report += '=' * 80 + '\n'
# Initialize the balance.
test_size = pred.size
balance = np.zeros_like(pred)
# Now go through the actual daily candles and figure out actual PL.
data = self.test_data[-test_size:]
for i in range(len(data)):
# Fetch the row.
row = data[i]
# Fetch the predicted and actual price change.
predicted = pred[i]
actual = util.get_price_change(row, self.pip_factor)
# Figure out the action we take and the units.
units = self.parse_units(predicted)
controls = self.parse_controls()
# Calculate profit/loss for this day.
p_l = util.get_profit_loss(row, units, **controls)
# Update daily profit/loss
balance[i] = balance[i - 1] + p_l
# Add to the report. row[0] is date.
report += util.format_row(row[0], units, predicted, actual, p_l)
# Add final total profit/loss to the report.
report += "Total profit/loss: {0}".format(balance[-1])
# Export the graphs if asked.
if 'export_plot' in kwargs:
# Do some plots.
common.plot(balance, kwargs['export_plot'])
# Print the report if asked.
if 'print_result' in kwargs and kwargs['print_result']:
print(report)
return balance
def get_best(self):
""" Produce a best instance of this strategy.
Args:
void.
Returns:
self: Euler instance. With the params and model having the
highest score among all combinations.
"""
# Log enter.
logger.info("Euler: Selecting best for %s.", self.instrument)
# Initialize the scores array and strategy parameters.
scores = []
strategy_params = util.get_euler_params()
# TODO: Do more than 1 run.
# Run for all predictive models.
counter = 0
for model in self.all_models:
scores_row = []
model_params = util.get_model_params(model)
# Try different model parameters.
for model_param in model_params:
scores_col = []
model = self.learner.build_model(model, 0.9, **model_param)
pred, _ = self.learner.test_model(model)
# And different strategy parameters, e.g. threshold.
for strategy_param in strategy_params:
self.set_params(**strategy_param)
# Do the dry run.
plot_name = '{0}_{1}.png'.format(self.instrument, counter)
balance = self.dry_run(pred, export_plot=plot_name)
# Determine the quality of the params via score.
scores_col.append(util.get_strategy_score(balance))
counter += 1
scores_row.append(scores_col)
scores.append(scores_row)
# Get the best and set the parameters to the best.
best = np.unravel_index(np.argmax(scores), np.array(scores).shape)
model = self.all_models[best[0]]
model_param = util.get_model_params(model)[best[1]]
model = self.learner.build_model(model, 1, **model_param)
logger.info("Best score is: %s.", str(np.array(scores)[best]))
self.set_params(**strategy_params[best[2]])
self.model = model
return self
def serialize(self):
""" Serialize this strategy to the designated location.
Args:
void.
Returns:
void.
"""
# Get the designated locations.
model_loc, param_loc = common.get_strategy_loc(self.instrument)
# Dump the data.
joblib.dump(self.model, model_loc)
json.dump(self.params, open(param_loc, 'w'))
return