def run(self):
filenames = self.gen_find()
if self._data_mode == 'Insample':
start_time = '2016-7-1 09:00:00.0'
end_time = '2016-10-1 09:00:00.0'
else:
start_time = '2016-10-1 09:00:00.0'
end_time = '2016-12-31 23:59:59.0'
for filename in filenames:
data = pd.read_csv(self._data_dir + '/' + filename, index_col=0)
data = data[(pd.to_datetime(data.index) >= start_time)
& (pd.to_datetime(data.index) < end_time)]
if len(data) == 0:
continue
else:
signals = [
SLMStrategy(data, self._slm, m).generatingsignal()
for m in np.arange(1, self._max_order + 1)
]
tcas = [self._tca for m in np.arange(1, self._max_order + 1)]
opts = [self._opt for m in np.arange(1, self._max_order + 1)]
validator_ensemble = ensembler(vectorizedbacktest, signals,
tcas, opts)
validator_ensemble.build()
validator_ensemble.run()
performance = validator_ensemble.calperformance()
performance.to_csv(self._valid_dir + '/performance_' +
filename)
validator_ensemble.plot()
plt.savefig(self._valid_dir + '/performance_' +
re.sub('.csv', '.png', filename))
plt.close()
def build(self,
model_order=8,
freq='5min',
start='20161001',
end='20161221',
offset=0,
tca=None,
fixed_cost=0.0002):
"""
Read in Data and the probability table
Args:
model_order: int or list. Number of model orders
"""
self.test_data = self.test_data[(self.test_data.index >= start)
& (self.test_data.index < end)]
self.signals = self.compile_signal(self.test_data,
self.prob_table,
model_orders=model_order)
self.ensemble = ensembler(vectorizedbacktest,
self.signals,
price=self.test_price,
tcas=[tca] * (model_order - 1),
fixed_cost=fixed_cost)
self.ensemble.build()
def build(self,
commodity,
exp_date,
model_order=7,
freq='5min',
offset=0,
start='20160701',
end='20161031',
tca=None):
self.test_data = self.compile_data(commodity,
exp_date,
start=start,
end=end)
self.signals = self.compile_signal(self.test_data,
self.prob_table,
max_order=model_order)
self.ensemble = ensembler(vectorizedbacktest, self.signals, tcas=tca)
self.ensemble.build()
def build(self,
commodity,
exp_date,
model_order=4,
freq=['{}min' for i in range(1, 5)],
offset=0,
start='20160701',
end='20161031',
tca=None):
self.test_data = self.compile_data(commodity,
exp_date,
freq=freq,
start=start,
end=end)
self.signals = self.compile_signal(self.test_data, self.prob_table,
model_order)
self.ensemble = ensembler(vectorizedbacktest,
self.signals,
tcas=tca,
labels=freq)
self.ensemble.build()
def build(self,
model_order=7,
freq='5min',
start='20161001',
end='20161221',
offset=0):
"""
Read in Data and the probability table
Args:
model_order: int or list. Number of model orders
"""
self.test_data = self.test_data[(self.test_data.index >= start)
& (self.test_data.index < end)]
self.signals = self.compile_signal(self.test_data,
self.prob_table,
model_orders=model_order)
self.ensemble = ensembler(vectorizedbacktest,
self.signals,
tcas=self._tca,
window=self._rollback_window)
self.ensemble.build()
def build(self,
model_order=7,
freq='5min',
start='20161001',
end='20161221',
offset=0,
tca=None,
labels=None):
"""
Read in Data and the probability table
Args:
model_order: int or list. Number of model orders
"""
self.test_data = self.test_data[(self.test_data.index >= start)
& (self.test_data.index < end)]
self.signals = self.compile_signal(self.test_data,
self.prob_table,
price=self.price,
model_orders=model_order)
all_signals = pd.concat(
[df[['signal', 'max_pct']] for df in self.signals],
axis=0,
keys=range(1, model_order + 1))
all_signals = all_signals.fillna(0)
all_signals = all_signals.unstack(level=0)
max_pct_signal = self.signals[0].copy()
max_pct_signal['signal'] = all_signals.apply(
lambda x: x.signal[x.max_pct.idxmax(axis=1)], axis=1)
self.signals.append(max_pct_signal)
self.ensemble = ensembler(vectorizedbacktest,
self.signals,
tcas=tca,
labels=labels,
price=self.price)
self.ensemble.build()
def run(self):
filenames = self.gen_find()
for filename in filenames:
data = pd.read_csv(self._data_dir + '/' + filename, index_col=0)
data = data[(pd.to_datetime(data.index) >= '2016-7-1 09:00:00.0') &
(pd.to_datetime(data.index) < '2016-10-1 09:00:00.0')]
if len(data) == 0:
continue
else:
signals = [
SLMStrategy(data, self._slm, m).generatingsignal()
for m in np.arange(1, self._max_order + 1)
]
validator_ensemble = ensembler(vectorizedbacktest, signals)
validator_ensemble.build()
validator_ensemble.run()
performance = validator_ensemble.calperformance()
performance.to_csv(self._valid_dir + '/performance_' +
filename)
validator_ensemble.plot()
plt.savefig(self._valid_dir + '/performance_' +
re.sub('.csv', '.png', filename))
plt.close()
# validator_ensemble.plot(target_col="benchmark")
# plt.savefig(self._valid_dir + '/benchmark_' + re.sub('.csv', '.png', filename))
# plt.close()
fig = plt.figure()
#benchmark = validator_ensemble.results[0]['return'].cumsum() + 1
benchmark = validator_ensemble.results[0]['benchmark']
#print(len(benchmark.index))
#initial_value = benchmark[0]
#benchmark = benchmark/initial_value
benchmark.plot()
plt.title('Benchmark')
fig.savefig(self._valid_dir + '/benchmark_' +
re.sub('.csv', '.png', filename))
plt.close()
if self._offsets_average:
if not self._average_return:
self._average_return = [
df['strategy'] for df in validator_ensemble.results
]
average_performance = performance
average_benchmark = validator_ensemble.results[0][
'LastPrice']
else:
self._average_return = [
df1.add(df2['strategy'], fill_value=0)
for (df1, df2) in zip(self._average_return,
validator_ensemble.results)
]
average_performance = average_performance.add(
performance, fill_value=0)
average_benchmark = average_benchmark.add(
validator_ensemble.results[0]['LastPrice'],
fill_value=0)
'''
average return of all offsets
'''
if self._offsets_average and (self._average_return is not None):
self._average_return = [(1 + df.divide(self._n_offsets)).cumprod()
for df in self._average_return]
average_benchmark = average_benchmark / average_benchmark[0]
average_benchmark = average_benchmark.to_frame()
average_benchmark['Date'] = pd.to_datetime(average_benchmark.index)
#print(average_benchmark)
fig = plt.figure()
plt.plot(average_benchmark.Date,
average_benchmark.LastPrice,
label='benchmark')
for avg_return, label in zip(
self._average_return,
np.arange(2, 2 + len(self._average_return))):
#avg_return.plot(label=label)
df = avg_return.to_frame()
df['Date'] = pd.to_datetime(df.index)
plt.plot(df.Date, df.strategy, label=label)
fig.autofmt_xdate()
plt.legend(loc='upper left')
plt.title('Equity Curve')
plt.show()
fig.savefig(self._valid_dir + '/performance_' + self._symbol +
'.png')
plt.close()
average_performance = average_performance.divide(self._n_offsets)
print(average_performance)
average_performance.to_csv(self._valid_dir + '/performance_' +
self._symbol + '.csv')
