o %ݫi@s,dZddlZddlZGdddejjZdS)z:K-means implementation. Authors * Luca Della Libera 2024 NcsveZdZdZfddZdfdd ZddZd d Zd d Zd dZ ddZ ddZ e ddZ e ddZZS)MiniBatchKMeansSklearnaA wrapper for scikit-learn MiniBatchKMeans, providing integration with PyTorch tensors. See https://scikit-learn.org/stable/modules/generated/sklearn.cluster.MiniBatchKMeans.html. Arguments --------- *args : tuple Positional arguments passed to scikit-learn `MiniBatchKMeans`. **kwargs : dict Keyword arguments passed to scikit-learn `MiniBatchKMeans`. Example ------- >>> import torch >>> device = "cpu" >>> n_clusters = 20 >>> batch_size = 8 >>> seq_length = 100 >>> hidden_size = 256 >>> model = MiniBatchKMeansSklearn(n_clusters).to(device) >>> input = torch.randn(batch_size, seq_length, hidden_size, device=device) >>> model.partial_fit(input) >>> labels = model(input) >>> labels.shape torch.Size([8, 100]) >>> centers = model.cluster_centers >>> centers.shape torch.Size([20, 256]) >>> len(list(model.buffers())) 1 >>> model.n_steps 1 >>> inertia = model.inertia(input) cspzddlm}Wntyd}|d7}t|wt||i||_td|_|jd|j dddS) Nr)MiniBatchKMeanszMThe optional dependency `scikit-learn` must be installed to use this module. z*Install using `pip install scikit-learn`. cpucluster_centersF) persistent) sklearn.clusterr ImportErrorsuper__init__kmeanstorchdeviceregister_buffercluster_centers_)selfargskwargsrerr_msg __class__S/home/ubuntu/.local/lib/python3.10/site-packages/speechbrain/lobes/models/kmeans.pyr /s    zMiniBatchKMeansSklearn.__init__Nc s||_t|S)z*See documentation of `torch.nn.Module.to`.)r r to)rr rrrrr>s zMiniBatchKMeansSklearn.tocCst|j|dS)zSaves the model to the specified file. Arguments --------- path : str The file path to save the model. N)joblibdumpr )rpathrrrsaveCszMiniBatchKMeansSklearn.savecCst||_|j|_dS)aLoads the model from the specified file. Arguments --------- path : str The file path from which to load the model. end_of_epoch : bool Indicates if this load is triggered at the end of an epoch. N)rloadr rr)rr end_of_epochrrrrMs zMiniBatchKMeansSklearn.loadcC0|jdd}|j||j|_dS)zFits the model to the input data. Arguments --------- input : torch.Tensor The input data tensor of shape (..., n_features). end_dimN)detachflattenrnumpyr fitrrrinput numpy_inputrrrr&Z  zMiniBatchKMeansSklearn.fitcCr)zPerforms an incremental fit of the model on the input data. Arguments --------- input : torch.Tensor The input data tensor of shape (..., n_features). r r!N)r#r$rr%r partial_fitrrr'rrrr+fr*z"MiniBatchKMeansSklearn.partial_fitcCsP|jdd}|j|}tj||jd }| |j dd}|S)a"Predicts cluster indices for the input data. Arguments --------- input : torch.Tensor The input data tensor of shape (..., n_features). Returns ------- torch.Tensor Predicted cluster indices of shape (...,). r r!r N) r#r$rr%r predictr tensorr longreshapeshape)rr(r) cluster_idxesrrrforwardrs  zMiniBatchKMeansSklearn.forwardcCs>|jdd}|j|}tj||jd }|S)a,Returns the inertia of the clustering. Arguments --------- input : torch.Tensor The input data tensor of shape (..., n_features). Returns ------- torch.Tensor Inertia (sum of squared distances to the cluster centers). r r!r,) r#r$rr%r scorer r/r float)rr(r)r5inertiarrrr7s zMiniBatchKMeansSklearn.inertiacCs|jjS)zReturns the number of minibatches processed. Returns ------- int Number of minibatches processed. )r n_steps_)rrrrn_stepss zMiniBatchKMeansSklearn.n_stepscCs@t|jdr|jj}tj||jd}|Stjd|jd}|S)zReturns the cluster centers. Returns ------- torch.Tensor Cluster centers of shape (n_clusters, n_features). rr,g)hasattrr rr r/r r6)rrrrrrs z'MiniBatchKMeansSklearn.cluster_centers_)N)__name__ __module__ __qualname____doc__r rrrr&r+r4r7propertyr9r __classcell__rrrrr s #      r)r>rr nnModulerrrrrs