def test_raise_long_history(org):
gen = copy.deepcopy(org.genotype)
gen["Data"]["history_days"] = 2000
phenotype = gen.get_phenotype()
net = model.Model(tuple(org._data.tickers), org._data.date, phenotype,
None)
with pytest.raises(model.ModelError) as error:
# noinspection PyStatementEffect
net.llh
assert issubclass(error.type, model.TooLongHistoryError)
def test_raise_gradient_error(org):
gen = copy.deepcopy(org.genotype)
gen["Scheduler"]["epochs"] /= 10
gen["Scheduler"]["max_lr"] = 10
phenotype = gen.get_phenotype()
net = model.Model(tuple(org._data.tickers), org._data.date, phenotype, None)
with pytest.raises(model.ModelError) as error:
# noinspection PyStatementEffect
net.llh
assert issubclass(error.type, model.GradientsError)
def test_llh_from_trained_and_reloaded_model(org):
gen = copy.deepcopy(org.genotype)
# Для ускорения обучения
gen["Scheduler"]["epochs"] /= 10
phenotype = gen.get_phenotype()
net = model.Model(tuple(org._data.tickers), org._data.date, phenotype, None)
assert bytes(net) == bytes()
llh = net.llh
pickled_model = bytes(net)
net = model.Model(tuple(org._data.tickers), org._data.date, phenotype, pickled_model)
assert llh == net.llh
assert bytes(net) == pickled_model
# Из кеша
assert llh == net.llh
assert llh == net._eval_llh()
def test_forecast(org):
gen = copy.deepcopy(org.genotype)
gen["Scheduler"]["epochs"] /= 10
phenotype = gen.get_phenotype()
net = model.Model(tuple(org._data.tickers), org._data.date, phenotype, org._data.model)
forecast = net.forecast()
assert isinstance(forecast, Forecast)
assert forecast.tickers == tuple(org._data.tickers)
assert forecast.date == org._data.date
assert forecast.history_days == phenotype["data"]["history_days"]
assert isinstance(forecast.mean, pd.Series)
assert forecast.mean.index.tolist() == list(org._data.tickers)
assert isinstance(forecast.std, pd.Series)
assert forecast.std.index.tolist() == list(org._data.tickers)