a  df@sddlmZmZmZddlZddlmZddlmZddlm Z e eee e fddddZ eed d d Zeeed d dZeed ddZee e eedddZe e eeeefdddZee e eeeeefdddZd"eeedee eeefdddZe ddd d!ZdS)#)OptionalTupleUnionN)Tensor)Literalrank_zero_warn) nan_strategynan_replace_valuereturncCs>|dvrtd||dkr:t|ttfs:td|dS)NreplaceZdropzPArgument `nan_strategy` is expected to be one of `['replace', 'drop']`, but got r zlArgument `nan_replace` is expected to be of a type `int` or `float` when `nan_strategy = 'replace`, but got ) ValueError isinstanceintfloat)r r rn/var/www/html/stable-diffusion-webui/venv/lib/python3.9/site-packages/torchmetrics/functional/nominal/utils.py_nominal_input_validationsr)confmatr cCs,|d|d}}td|||S)zDCompute the expected frequenceis from the provided confusion matrix.rz r, c -> rc)sumtorchZeinsum)rZmargin_sum_rowsZmargin_sum_colsrrr_compute_expected_freqs#sr)rbias_correctionr cCst|}|t|j|jd}|dkr>> import torch >>> from torchmetrics.functional.nominal.utils import _drop_empty_rows_and_cols >>> _ = torch.manual_seed(22) >>> matrix = torch.randint(10, size=(3, 3)) >>> matrix[1, :] = matrix[:, 1] = 0 >>> matrix tensor([[9, 0, 6], [0, 0, 0], [2, 0, 8]]) >>> _drop_empty_rows_and_cols(matrix) tensor([[9, 6], [2, 8]]) rrN)r)rrrr_drop_empty_rows_and_cols=sr') phi_squaredn_rowsn_cols confmat_sumr cCs0ttjd|jd||d|d|dS)z#Compute bias-corrected Phi Squared.rrr)rmaxr!r)r(r)r*r+rrr_compute_phi_squared_correctedSs(r-)r)r*r+r cCs8||dd|d}||dd|d}||fS)z2Compute bias-corrected number of rows and columns.rrr)r)r*r+rows_correctedcols_correctedrrr _compute_rows_and_cols_corrected_sr0cCs(t||||}t|||\}}|||fS)zBCompute bias-corrected Phi Squared and number of rows and columns.)r-r0)r(r)r*r+Zphi_squared_correctedr.r/rrr_compute_bias_corrected_valuesfsr1r rr )predstargetr r r cCsD|dkr||||fSt||}||||fS)a0Handle ``NaN`` values in input data. If ``nan_strategy = 'replace'``, all ``NaN`` values are replaced with ``nan_replace_value``. If ``nan_strategy = 'drop'``, all rows containing ``NaN`` in any of two vectors are dropped. Args: preds: 1D tensor of categorical (nominal) data target: 1D tensor of categorical (nominal) data nan_strategy: Indication of whether to replace or drop ``NaN`` values nan_replace_value: Value to replace ``NaN`s when ``nan_strategy = 'replace``` Returns: Updated ``preds`` and ``target`` tensors which contain no ``Nan`` Raises: ValueError: If ``nan_strategy`` is not from ``['replace', 'drop']``. ValueError: If ``nan_strategy = replace`` and ``nan_replace_value`` is not of a type ``int`` or ``float``. r )Z nan_to_numr logical_orisnan)r2r3r r Zrows_contain_nanrrr_handle_nan_in_dataosr6) metric_namer cCstd|ddS)NzUnable to compute zG using bias correction. Please consider to set `bias_correction=False`.r)r7rrr&_unable_to_use_bias_correction_warnings r8)r r)typingrrrrrZtyping_extensionsrZtorchmetrics.utilities.printsrstrrrrrboolr&r'r-r0r1r6r8rrrrs8