o .wij' @sRddlmZmZddlZddlmZmZddlmZddlm Z d,dede d efd d Z d-d edd dfddZ ded efddZ dedeeedfd efddZ d.dededee ded ef ddZded efddZdeded dfdd Zd!ed"ed dfd#d$Zd%ed&ed dfd'd(Z   d/deedeed)eed efd*d+ZdS)0)OptionalUnionN)Tensortensor)Literal)_check_same_shapeh㈵>xatolreturncCst|dkt||kS)zReturn True if all elements of tensor are nonnegative within certain tolerance. Args: x: tensor atol: absolute tolerance Returns: Boolean tensor indicating if all values are nonnegative )torch logical_orabsall)r r re/home/ubuntu/sommelier/.venv/lib/python3.10/site-packages/torchmetrics/functional/clustering/utils.pyis_nonnegatives r arithmeticaverage_methodmin geometricrmaxcCs|dvr td|dS)NrzaExpected argument `average_method` to be one of `min`, `geometric`, `arithmetic`, `max`,but got  ValueError)rrrr_validate_average_method_arg%srcCst|dkr td|jdSttj|ddd}||dk}|dkr,td|jdS|}t||t|t| S)aCalculate entropy for a tensor of labels. Final calculation of entropy is performed in log form to account for roundoff error. Args: x: labels Returns: entropy: entropy of tensor Example: >>> from torchmetrics.functional.clustering.utils import calculate_entropy >>> labels = torch.tensor([1, 3, 2, 2, 1]) >>> calculate_entropy(labels) tensor(1.0549) r?)deviceTreturn_inverser ) lenrrr bincountuniquesizesumlog)r pnrrrcalculate_entropy/s   $r*r(cCst|s t|s tdt|tr<|dkr|S|dkr(tt| S|dkr0|S|dkr8| Stdtt ||d|S)atReturn generalized (power) mean of a tensor. Args: x: tensor p: power Returns: generalized_mean: generalized mean Example (p="min"): >>> from torchmetrics.functional.clustering.utils import calculate_generalized_mean >>> x = torch.tensor([1, 3, 2, 2, 1]) >>> calculate_generalized_mean(x, "min") tensor(1) Example (p="geometric"): >>> from torchmetrics.functional.clustering.utils import calculate_generalized_mean >>> x = torch.tensor([1, 3, 2, 2, 1]) >>> calculate_generalized_mean(x, "geometric") tensor(1.6438) z&`x` must contain positive real numbersrrrrz<'method' must be 'min', 'geometric', 'arirthmetic', or 'max'r) r is_complexrr isinstancestrrexpmeanr'rpow)r r(rrrcalculate_generalized_meanNs r1Fpredstargetepssparsec Cs|dur |dur td|jdks|jdkr#td|jd|jdtj|dd\}}tj|dd\}}|d }|d } tt||ftj|j d |j |j d | |f} |sd| } |rd| |} | S) arCalculate contingency matrix. Args: preds: predicted labels target: ground truth labels eps: value added to contingency matrix sparse: If True, returns contingency matrix as a sparse matrix. Else, return as dense matrix. `eps` must be `None` if `sparse` is `True`. Returns: contingency: contingency matrix of shape (n_classes_target, n_classes_preds) Example: >>> import torch >>> from torchmetrics.functional.clustering.utils import calculate_contingency_matrix >>> preds = torch.tensor([2, 1, 0, 1, 0]) >>> target = torch.tensor([0, 2, 1, 1, 0]) >>> calculate_contingency_matrix(preds, target, eps=1e-16) tensor([[1.0000e+00, 1.0000e-16, 1.0000e+00], [1.0000e+00, 1.0000e+00, 1.0000e-16], [1.0000e-16, 1.0000e+00, 1.0000e-16]]) NTz.Cannot specify `eps` and return sparse tensor.r!z)Expected 1d `preds` and `target` but got  and .rrdtyper) rndimdimr r$r%sparse_coo_tensorstackonesshaper9rto_dense) r2r3r4r5 preds_classes preds_idxtarget_classes target_idxnum_classes_predsnum_classes_target contingencyrrrcalculate_contingency_matrixws.   rHcCs2|jdkrtd|jdt|pt| S)z/Check if tensor of labels is real and discrete.r!z-Expected arguments to be 1-d tensors but got z -d tensors.)r:rr is_floating_pointr+)r rrr_is_real_discrete_labels rJcCs8t||t|r t|std|jd|jddS)zCheck shape of input tensors and if they are real, discrete tensors. Args: preds: predicted labels target: ground truth labels z2Expected real, discrete values for x but received r6r7N)rrJrr9)r2r3rrrcheck_cluster_labelss rKdatalabelscCsV|jdkrtd|jd|std|jd|jdkr)td|jdd S) zDValidate that the input data and labels have correct shape and type.zExpected 2D data, got zD data insteadz"Expected floating point data, got z data insteadr!zExpected 1D labels, got zD labels insteadN)r:rrIr9)rLrMrrr _validate_intrinsic_cluster_datas  rO num_labels num_samplescCs0d|kr |ksntd|d|ddS)z>> import torch >>> from torchmetrics.functional.clustering.utils import calculate_pair_cluster_confusion_matrix >>> preds = torch.tensor([0, 0, 1, 1]) >>> target = torch.tensor([1, 1, 0, 0]) >>> calculate_pair_cluster_confusion_matrix(preds, target) tensor([[8, 0], [0, 4]]) >>> preds = torch.tensor([0, 0, 1, 2]) >>> target = torch.tensor([0, 0, 1, 1]) >>> calculate_pair_cluster_confusion_matrix(preds, target) tensor([[8, 2], [0, 2]]) Nz:Must provide either `preds` and `target` or `contingency`.zDMust provide either `preds` and `target` or `contingency`, not both.zDMust provide `contingency` if `preds` and `target` are not provided.r!)r;rrNr8)r!r!)r!r)rr!)rr)rrHr&r zerosr9rT)r2r3rGrQsum_csum_k sum_squared pair_matrixrrr'calculate_pair_cluster_confusion_matrixs$.      rY)r)r)NF)NNN)typingrrr rrtyping_extensionsrtorchmetrics.utilities.checksrfloatrrr*intr1boolrHrJrKrOrRrYrrrrsT     "* 9