def test_with_tuning(mock_data):
with patch(THRESH_STR, return_value=(60, 20)) as mock_thresh:
train_thresh, test_thresh = calculate_split_thresholds(
mock_data, 0.603, 0.2015, True)
mock_thresh.assert_called_once_with(mock_data, 0.60, 0.20)
assert train_thresh == 60
assert test_thresh == 20
def test_not_tuning(mock_data):
with patch(THRESH_STR, return_value=(80, 20)) as mock_thresh:
train_thresh, test_thresh = calculate_split_thresholds(
mock_data, 0.6045, 0.2027, False)
mock_thresh.assert_called_once_with(mock_data, 0.80, 0.20)
assert train_thresh == 80
assert test_thresh == 20
def split_time_series_data(self, ts_data, tuning=True):
"""Splits `ts_data` into two sets for the modeling process.
If `tuning` is True, `ts_data` is split into a training and
validation set. Otherwise, it is split into a training + validation
and testing set.
Parameters
----------
ts_data: pd.Object
A pandas Object representing the data to be split.
tuning: bool
A boolean indicating whether the split is being done for training.
Returns
-------
pd.DataFrame, pd.DataFrame
Two pandas DataFrames where the first dataframe contains the
records in the training set and the second contains the records
in the testing set.
"""
train_thresh, test_thresh = utils.calculate_split_thresholds(
ts_data, self._train_prop, self._val_prop, tuning)
return ts_data[:train_thresh], ts_data[train_thresh:test_thresh]
def split_data(self, model_data, tuning=True):
"""Splits `model_data` into training and testing time series.
If `tuning` is True, `model_data` is split into training and
validation sets. Otherwise, it is split into training + validation
and testing sets.
Parameters
----------
model_data: np.array
A numpy array containing the data of the time series being
forecast.
tuning: bool
A boolean value indicating whether the split is for tuning.
Returns
-------
np.array, np.array
Two numpy arrays representing the records in the train and
test sets respectively.
"""
train_thresh, test_thresh = utils.calculate_split_thresholds(
model_data, self._train_prop, self._val_prop, tuning)
return model_data[:train_thresh], model_data[train_thresh:test_thresh]
def split_data(self, model_data, tuning=True):
"""Splits `model_data` into the training and testing set.
If `tuning` is True, `model_data` is split into training and
validation sets. Otherwise, it is split into training + validation
and testing sets.
Parameters
----------
model_data: pandas.Object
A pandas Object (Series or DataFrame) containing the data used to
build the trace model.
tuning: bool
A boolean value indicating whether the split is for tuning.
Returns
-------
pandas.Object, pandas.Object
The two pandas objects obtained from `model_data` after applying
the train-test split.
"""
train_thresh, test_thresh = utils.calculate_split_thresholds(
model_data, self._train_prop, self._val_prop, tuning)
return model_data[:train_thresh], model_data[train_thresh:test_thresh]
def split_data(self, model_data, tuning=True):
"""Splits `model_data` into training and testing sets.
Parameters
----------
model_data: pd.DataFrame
A pandas DataFrame containing the time series that are being
modeled, with a separate column for each time series that is
being modeled by the multivariate model.
tuning: bool
A boolean value indicating whether the split is for tuning.
Returns
-------
pd.DataFrame, pd.DataFrame
Two pandas DataFrames representing the records in the train and
test sets respectively.
"""
train_thresh, test_thresh = utils.calculate_split_thresholds(
model_data, self._train_prop, self._val_prop, tuning)
model_data = model_data.reset_index(drop=True)
return model_data[:train_thresh], model_data[train_thresh:test_thresh]
def get_total_spare(self, trace, tuning=True):
"""The spare amount of the target for `trace` over the test window.
Parameters
----------
trace: Trace
The `Trace` for which the spare is calculated.
tuning: bool
A boolean value indicating whether the spare is being calculated
to tune the model or evaluate the model on the test set.
Returns
-------
float
A float representing the total spare amount of the target
variable for `trace` over the test window.
"""
target_ts = trace.get_target_time_series(self._model_vars[0])
train_thresh, test_thresh = utils.calculate_split_thresholds(
target_ts, self._train_prop, self._val_prop, tuning)
return trace.get_spare_resource_in_window(
self._model_vars[0], train_thresh, test_thresh)