Performance metrics

torchid.metrics.r_squared(y_true, y_pred, time_axis=0)[source]

Computes the R-square index.

The R-squared index is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

r_squared_val – Array of r_squared value.

Return type:

np.array

torchid.metrics.rmse(y_true, y_pred, time_axis=0)[source]

Computes the Root Mean Square Error (RMSE).

The RMSE index is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

RMSE – Array of r_squared value.

Return type:

np.array

torchid.metrics.nrmse(y_true, y_pred, time_axis=0)[source]

Computes the Normalized Root Mean Square Error (NRMSE).

The NRMSE index is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

NRMSE – Array of r_squared value.

Return type:

np.array

torchid.metrics.error_mean(y_true, y_pred, time_axis=0)[source]

Computes the error mean value.

The error mean is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

e_mean – Array of error means.

Return type:

np.array

torchid.metrics.mae(y_true, y_pred, time_axis=0)[source]

Computes the error Mean Absolute Value (MAE)

The MAE index is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

e_mae – Array of error mean absolute values.

Return type:

np.array

torchid.metrics.fit_index(y_true, y_pred, time_axis=0)[source]

Computes the per-channel fit index.

The fit index is commonly used in System Identification. See the definition in the System Identification Toolbox or in the paper ‘Nonlinear System Identification: A User-Oriented Road Map’, https://arxiv.org/abs/1902.00683, page 31. The fit index is computed separately on each channel.

Parameters:

  • y_true (np.array) – Array of true values. If must be at least 2D.

  • y_pred (np.array) – Array of predicted values. If must be compatible with y_true’

  • time_axis (int) – Time axis. All other axes define separate channels.

Returns:

fit – Array of fit index.

Return type:

np.array