ts_benchmark.evaluation.metrics package

ts_benchmark.evaluation.metrics.regression_metrics module

Functions

mae(actual, predicted, **kwargs)	Mean Absolute Error
mae_norm(actual, predicted, scaler, **kwargs)	Mean Absolute Error
mape(actual, predicted, **kwargs)	Mean Absolute Percentage Error
mape_norm(actual, predicted, scaler, **kwargs)	Mean Absolute Percentage Error Properties: + Easy to interpret + Scale independent - Biased, not symmetric - Undefined when actual[t] == 0
mase(actual, predicted, hist_data[, seasonality])	Mean Absolute Scaled Error Baseline (benchmark) is computed with naive forecasting (shifted by @seasonality)
mase_norm(actual, predicted, scaler, hist_data)	Mean Absolute Scaled Error Baseline (benchmark) is computed with naive forecasting (shifted by @seasonality)
mse(actual, predicted, **kwargs)	Mean Squared Error
mse_norm(actual, predicted, scaler, **kwargs)	Mean Squared Error
msmape(actual, predicted[, epsilon])	Function to calculate series wise smape values
msmape_norm(actual, predicted, scaler[, epsilon])	Function to calculate series wise smape values
rmse(actual, predicted, **kwargs)	Root Mean Squared Error
rmse_norm(actual, predicted, scaler, **kwargs)	Root Mean Squared Error
smape(actual, predicted, **kwargs)	Symmetric Mean Absolute Percentage Error
smape_norm(actual, predicted, scaler, **kwargs)	Symmetric Mean Absolute Percentage Error
wape(actual, predicted, **kwargs)	Masked weighted absolute percentage error (WAPE)
wape_norm(actual, predicted, scaler, **kwargs)	Masked weighted absolute percentage error (WAPE)

mae(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Mean Absolute Error

mae_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Mean Absolute Error

mape(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Mean Absolute Percentage Error

Properties: + Easy to interpret + Scale independent - Biased, not symmetric - Undefined when actual[t] == 0

mape_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Mean Absolute Percentage Error Properties: + Easy to interpret + Scale independent - Biased, not symmetric - Undefined when actual[t] == 0

mase(actual: numpy.ndarray, predicted: numpy.ndarray, hist_data: numpy.ndarray, seasonality: int = 2, **kwargs)

Mean Absolute Scaled Error Baseline (benchmark) is computed with naive forecasting (shifted by @seasonality)

mase_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, hist_data: numpy.ndarray, seasonality: int = 2, **kwargs)

Mean Absolute Scaled Error Baseline (benchmark) is computed with naive forecasting (shifted by @seasonality)

mse(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Mean Squared Error

mse_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Mean Squared Error

msmape(actual: numpy.ndarray, predicted: numpy.ndarray, epsilon: float = 0.1, **kwargs)

Function to calculate series wise smape values

Parameters:

• actual – Array of actual values.

• predicted – Array of predicted values.

• epsilon – Small constant to avoid division by zero.

Returns:

Mean symmetric mean absolute percentage error (MSMAPE) as a percentage.

msmape_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, epsilon: float = 0.1, **kwargs)

Function to calculate series wise smape values

Parameters:

• actual – Array of actual values.

• predicted – Array of predicted values.

• scaler – Object used to scale the actual and predicted values.

• epsilon – Small constant to avoid division by zero.

Returns:

Mean symmetric mean absolute percentage error (MSMAPE) as a percentage.

rmse(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Root Mean Squared Error

rmse_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Root Mean Squared Error

smape(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Symmetric Mean Absolute Percentage Error

smape_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Symmetric Mean Absolute Percentage Error

wape(actual: numpy.ndarray, predicted: numpy.ndarray, **kwargs)

Masked weighted absolute percentage error (WAPE)

Parameters:

• predicted – Predicted values.

• actual – Ground truth labels.

Returns:

Masked mean absolute error.

wape_norm(actual: numpy.ndarray, predicted: numpy.ndarray, scaler: object, **kwargs)

Masked weighted absolute percentage error (WAPE)

Parameters:

• actual – Array of actual values.

• predicted – Array of predicted values.

• scaler – Object used to scale the actual and predicted values.

Returns:

Weighted absolute percentage error (WAPE) as a percentage.

ts_benchmark.evaluation.metrics.utils module

Functions

find_length(data)	Automatically calculate the appropriate period length for time series data.
get_list_anomaly(labels)	Get a list of anomaly interval lengths from time series labels.

find_length(data: numpy.ndarray) → int

Automatically calculate the appropriate period length for time series data.

Parameters:

data – Time series data.

Returns:

The automatically calculated period length.

get_list_anomaly(labels: numpy.ndarray) → List[int]

Get a list of anomaly interval lengths from time series labels.

Parameters:

labels – A list of time series labels, where 1 indicates anomaly and 0 indicates normal.

Returns:

A list of anomaly interval lengths.