spared.metrics.get_metrics

spared.metrics.get_metrics(gt_mat: Union[array, Tensor], pred_mat: Union[array, Tensor], mask: Union[array, Tensor], detailed: bool = False) → dict[source]

Get general regression metrics

This function receives 2 matrices of shapes (n_samples, n_genes) and computes the following metrics:

  • Pearson correlation (gene-wise) [PCC-Gene]

  • Pearson correlation (patch-wise) [PCC-Patch]

  • r2 score (gene-wise) [R2-Gene]

  • r2 score (patch-wise) [R2-Patch]

  • Mean squared error [MSE]

  • Mean absolute error [MAE]

  • Global metric [Global] (Global = PCC-Gene + R2-Gene + PCC-Patch + R2-Patch - MAE - MSE)

If detailed == True. Then the function returns these aditional keys (all of them are numpy arrays):

  • Individual pearson correlation for every gene [PPC-Gene-detailed]

  • Individual pearson correlation for every patch [PPC-Patch-detailed]

  • Individual r2 score for every gene [R2-Gene-detailed]

  • Individual r2 score for every patch [R2-Gene-detailed]

  • Individual MSE for every gene [detailed_mse_gene]

  • Individual MAE for every gene [detailed_mae_gene]

  • Individual average error for every gene [detailed_error_gene]

  • Flat concatenation of all errors in valid positions [detailed_errors]

Parameters:

  • gt_mat (Union[np.array, torch.Tensor]) – Ground truth matrix of shape (n_samples, n_genes).

  • pred_mat (Union[np.array, torch.Tensor]) – Predicted matrix of shape (n_samples, n_genes).

  • mask (Union[np.array, torch.Tensor]) – Boolean mask with False in positions that must be ignored in metric computation (n_samples, n_genes).

  • detailed (bool) – If True, the dictionary also includes the detailed metrics.

Returns:

Dictionary containing the metrics computed. The keys are: [‘PCC-Gene’, ‘PCC-Patch’, ‘R2-Gene’, ‘R2-Patch’, ‘MSE’, ‘MAE’, ‘Global’]

Return type:

dict