def create_preprocessing_pipeline(power_scaler):
pipeline = Pipeline(path="../results/preprocessing")
# Deal with missing values through linear interpolation
imputer_power_statistics = LinearInterpolater(method="nearest", dim="time",
name="imputer_power")(x=pipeline["scaler_power"])
# Scale the data using a standard SKLearn scaler
scale_power_statistics = power_scaler(x=imputer_power_statistics)
# Create lagged time series to later be used in the regression
ClockShift(lag=1)(x=scale_power_statistics)
ClockShift(lag=2)(x=scale_power_statistics)
return pipeline
class TestClockShift(unittest.TestCase):
def setUp(self):
self.clock_shift = ClockShift(lag=2)
def tearDown(self):
self.clock_shift = None
def test_get_params(self):
params = self.clock_shift.get_params()
self.assertEqual({"lag": 2, "indexes": None}, params)
def test_set_params(self):
params = self.clock_shift.get_params()
self.assertEqual({"lag": 2, "indexes": None}, params)
self.clock_shift.set_params(indexes=["time"], lag=24)
params = self.clock_shift.get_params()
self.assertEqual({"lag": 24, "indexes": ["time"]}, params)
def test_transform(self):
time = pd.date_range('2000-01-01', freq='24H', periods=7)
ds = xr.Dataset({'foo': ('time', [2, 3, 4, 5, 6, 7, 8]), 'time': time})
result = self.clock_shift.transform(ds)
time = pd.date_range('2000-01-01', freq='24H', periods=7)
expected_result = xr.Dataset({
'foo': ('time', [np.nan, np.nan, 2., 3., 4.5, 5., 6.]),
'time':
time
})
self.assertEqual(result.values(), expected_result)
def test_transform_exception(self):
time = pd.date_range('2000-01-01', freq='24H', periods=7)
self.clock_shift.set_params(indexes=["FOO"])
ds = xr.Dataset({'foo': ('time', [2, 3, 4, 5, 6, 7, 8]), 'time': time})
with self.assertRaises(WrongParameterException) as context:
self.clock_shift.transform(ds)
self.assertEqual(
context.exception.message,
"Not all indexes (['FOO']) are in the indexes of x (['time']). "
"Perhaps you set the wrong indexes with set_params or during the initialization of the ClockShift."
)
def pipe(params):
keras_model = get_keras_model(params)
pipeline = Pipeline(path="../results")
imputer_power_statistics = LinearInterpolater(
method='nearest', dim='time',
name='imputer_power')(x=pipeline['load_power_statistics'])
power_scaler = SKLearnWrapper(module=StandardScaler(),
name='scaler_power')
scale_power_statistics = power_scaler(x=imputer_power_statistics)
shift_power_statistics = ClockShift(
lag=1, name='ClockShift_Lag1')(x=scale_power_statistics)
shift_power_statistics2 = ClockShift(
lag=2, name='ClockShift_Lag2')(x=scale_power_statistics)
keras_wrapper = KerasWrapper(keras_model,
fit_kwargs={'batch_size': 32, 'epochs': 100, 'verbose': 0},
compile_kwargs={'loss': 'mse', 'optimizer': 'Adam', 'metrics': ['mse']}) \
(ClockShift_Lag1=shift_power_statistics,
ClockShift_Lag2=shift_power_statistics2,
target=scale_power_statistics)
inverse_power_scale_dl = power_scaler(
x=keras_wrapper,
computation_mode=ComputationMode.Transform,
use_inverse_transform=True,
callbacks=[LinePlotCallback('prediction')])
rmse_dl = RmseCalculator()(keras_model=inverse_power_scale_dl,
y=pipeline['load_power_statistics'],
callbacks=[CSVCallback('RMSE')])
pipeline.train(train)
result = pipeline.test(test)
return {
"loss": float(result['RmseCalculator'].values),
"status": STATUS_OK,
"eval_time": time.time() - start
}
CalendarFeature.month,
CalendarFeature.weekday,
CalendarFeature.weekend
])(x=pipeline["load_power_statistics"])
# Deal with missing values through linear interpolation
imputer_power_statistics = LinearInterpolater(
method="nearest", dim="time",
name="imputer_power")(x=pipeline["load_power_statistics"])
# Scale the data using a standard SKLearn scaler
power_scaler = SKLearnWrapper(module=StandardScaler(), name="scaler_power")
scale_power_statistics = power_scaler(x=imputer_power_statistics)
# Create lagged time series to later be used in the regression
shift_power_statistics = ClockShift(
lag=1, name="ClockShift_Lag1")(x=scale_power_statistics)
shift_power_statistics2 = ClockShift(
lag=2, name="ClockShift_Lag2")(x=scale_power_statistics)
# Create a linear regression that uses the lagged values to predict the current value
# NOTE: SKLearnWrapper has to collect all **kwargs itself and fit it against target.
# It is also possible to implement a join/collect class
regressor_power_statistics = SKLearnWrapper(module=LinearRegression(
fit_intercept=True))(
power_lag1=shift_power_statistics,
power_lag2=shift_power_statistics2,
calendar=calendar,
target=scale_power_statistics,
callbacks=[LinePlotCallback('linear_regression')],
)
def setUp(self):
self.clock_shift = ClockShift(lag=2)