o X۷iS@sddlmZddlZdZdZz ddlmZdZWn!ey9z ddlmZdZWn ey6dZdZYnwYnwddZ dd Z Gd d d Z e d d hd e dhdd e dhdd e dddhd e dddhd e dhdddgde dhdd e dddhd e dhd d e d!d!hdgde d"hd#d e d$d$hd e d%hd&d g Z d'd(e DZ ed)d*e DZee Zd+d,Zd-dZd.d Zd/dZd0dZd1dZd2dZd3dZd4dZd5dZd6d!Zd7d"Zd8d$Zd9d%Z dBd:d;Z!dCdd<d=d>Z"dDd@dAZ#dS)E) annotationsNF)pairwise_distanceTcCstj||dS)Ndtype)cupyascontiguousarray)Xout_typer R/home/ubuntu/vllm_env/lib/python3.10/site-packages/cupyx/scipy/spatial/distance.py_convert_to_typesr cKs\|j}|j|vr|||jn|d}t||d}|j}|r)||||fi|}|||fS)Nr)r )typesrindexr validator)rmn metric_infokwargsr typ_validate_kwargsr r r _validate_pdist_inputs"  rc@seZdZ  dddZdS) MetricInfoNcCs||_||_||_||_dSN)canonical_name_aka_ validator_types_)selfcanonical_nameakarr r r r __init__)s zMetricInfo.__init__)NNNN)__name__ __module__ __qualname__r r r r r r'srcanberra)rr chebyshev>chchebcheby chebychevr% cityblock>ccbcblockr* correlationcocosinecoshamming>hhahammr2matchingdoublebool)rrr euclidean>eeueuclidr9 jensenshannonjs minkowski>rmipnormr? russellrao sqeuclidean>sqesqeuclidrC hellinger kl_divergence>kldkl_divrGcCsi|]}|j|qSr )r).0infor r r lsrLccs$|] }|jD]}||fVqqdSr)r)rJrKaliasr r r msrNcCststs tddSdS)NzJcuVS >= 24.12 or pylibraft < 24.12 should be installed to use this feature)cuvs_availablepylibraft_available RuntimeErrorr r r r check_soft_dependenciests rRcCsltt|r dnd}t|rdnd}||kr!td||ftjd|j|d}t|||d||dS)aCompute the Minkowski distance between two 1-D arrays. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) p (float): The order of the norm of the difference :math:`{\|u-v\|}_p`. Note that for :math:`0 < p < 1`, the triangle inequality only holds with an additional multiplicative factor, i.e. it is only a quasi-metric. Returns: minkowski (double): The Minkowski distance between vectors `u` and `v`. FC9u and v must have the same layout (u.order=%s, v.order=%srWrorderr?rrrRr isfortran ValueErroremptyrr)uvpu_orderv_order output_arrr r r r?{scCjtt|r dnd}t|rdnd}||kr!td||ftjd|j|d}t|||d|dS)azCompute the Canberra distance between two 1-D arrays. The Canberra distance is defined as .. math:: d(u, v) = \sum_{i} \frac{| u_i - v_i |}{|u_i| + |v_i|} Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: canberra (double): The Canberra distance between vectors `u` and `v`. rSrTrUrVrXr$rZr[r_r`rbrcrdr r r r$cCre)afCompute the Chebyshev distance between two 1-D arrays. The Chebyshev distance is defined as .. math:: d(u, v) = \max_{i} |u_i - v_i| Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: chebyshev (double): The Chebyshev distance between vectors `u` and `v`. rSrTrUrVrXr%rZr[rfr r r r%rgcCre)a}Compute the City Block (Manhattan) distance between two 1-D arrays. The City Block distance is defined as .. math:: d(u, v) = \sum_{i} |u_i - v_i| Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: cityblock (double): The City Block distance between vectors `u` and `v`. rSrTrUrVrXr*rZr[rfr r r r*cCre)a2Compute the correlation distance between two 1-D arrays. The correlation distance is defined as .. math:: d(u, v) = 1 - \frac{(u - \bar{u}) \cdot (v - \bar{v})}{ \|(u - \bar{u})\|_2 \|(v - \bar{v})\|_2} where :math:`\bar{u}` is the mean of the elements of :math:`u` and :math:`x \cdot y` is the dot product. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: correlation (double): The correlation distance between vectors `u` and `v`. rSrTrUrVrXr.rZr[rfr r r r.scCre)aCompute the Cosine distance between two 1-D arrays. The Cosine distance is defined as .. math:: d(u, v) = 1 - \frac{u \cdot v}{\|u\|_2 \|v\|_2} where :math:`x \cdot y` is the dot product. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: cosine (double): The Cosine distance between vectors `u` and `v`. rSrTrUrVrXr0rZr[rfr r r r0cCre)a/Compute the Hamming distance between two 1-D arrays. The Hamming distance is defined as the proportion of elements in both `u` and `v` that are not in the exact same position: .. math:: d(u, v) = \frac{1}{n} \sum_{k=0}^n u_i \neq v_i where :math:`x \neq y` is one if :math:`x` is different from :math:`y` and zero otherwise. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: hamming (double): The Hamming distance between vectors `u` and `v`. rSrTrUrVrXr2rZr[rfr r r r2=cCre)aCompute the Euclidean distance between two 1-D arrays. The Euclidean distance is defined as .. math:: d(u, v) = \left(\sum_{i} (u_i - v_i)^2\right)^{\sfrac{1}{2}} Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: euclidean (double): The Euclidean distance between vectors `u` and `v`. rSrTrUrVrXr9rZr[rfr r r r9`rgcCre)aCompute the Jensen-Shannon distance between two 1-D arrays. The Jensen-Shannon distance is defined as .. math:: d(u, v) = \sqrt{\frac{KL(u \| m) + KL(v \| m)}{2}} where :math:`KL` is the Kullback-Leibler divergence and :math:`m` is the pointwise mean of `u` and `v`. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: jensenshannon (double): The Jensen-Shannon distance between vectors `u` and `v`. rSrTrUrVrXr=rZr[rfr r r r=rjcCre)a-Compute the Russell-Rao distance between two 1-D arrays. The Russell-Rao distance is defined as the proportion of elements in both `u` and `v` that are in the exact same position: .. math:: d(u, v) = \frac{1}{n} \sum_{k=0}^n u_i = v_i where :math:`x = y` is one if :math:`x` is different from :math:`y` and zero otherwise. Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: hamming (double): The Hamming distance between vectors `u` and `v`. rSrTrUrVrXrBrZr[rfr r r rBrjcCre)aCompute the squared Euclidean distance between two 1-D arrays. The squared Euclidean distance is defined as .. math:: d(u, v) = \sum_{i} (u_i - v_i)^2 Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: sqeuclidean (double): The squared Euclidean distance between vectors `u` and `v`. rSrTrUrVrXrCrZr[rfr r r rCrhcCre)aCompute the Hellinger distance between two 1-D arrays. The Hellinger distance is defined as .. math:: d(u, v) = \frac{1}{\sqrt{2}} \sqrt{ \sum_{i} (\sqrt{u_i} - \sqrt{v_i})^2} Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: hellinger (double): The Hellinger distance between vectors `u` and `v`. rSrTrUrVrXrFrZr[rfr r r rFricCs~tt|}t|}t|rdnd}t|rdnd}||kr+td||ftjd|j|d}t|||d|dS)aCompute the Kullback-Leibler divergence between two 1-D arrays. The Kullback-Leibler divergence is defined as .. math:: KL(U \| V) = \sum_{i} U_i \log{\left(\frac{U_i}{V_i}\right)} Args: u (array_like): Input array of size (N,) v (array_like): Input array of size (N,) Returns: kl_divergence (double): The Kullback-Leibler divergence between vectors `u` and `v`. rSrTrUrVrXrGrZ)rRrasarrayr\r]r^rrrfr r r rGs  cKstts tr|jdvrtj|dd}tstr#|jdvr#tj|dd}t|r*dnd}t|r3dnd}||krAtd||f|j}|j}t |dkrQtdt |dkr[td |d |d krgtd |d } |d } d |vrw|d nd} |durtr|jdkstr|jdvrt|rdnd} | |krtd| |j ddd}|j| | fkrt || | ft |t r|} t| d}|dur|dur|n tj| | f|j|d}t||||| |Std| td)aCompute distance between each pair of the two collections of inputs. Args: XA (array_like): An :math:`m_A` by :math:`n` array of :math:`m_A` original observations in an :math:`n`-dimensional space. Inputs are converted to float type. XB (array_like): An :math:`m_B` by :math:`n` array of :math:`m_B` original observations in an :math:`n`-dimensional space. Inputs are converted to float type. metric (str, optional): The distance metric to use. The distance function can be 'canberra', 'chebyshev', 'cityblock', 'correlation', 'cosine', 'euclidean', 'hamming', 'hellinger', 'jensenshannon', 'kl_divergence', 'matching', 'minkowski', 'russellrao', 'sqeuclidean'. out (cupy.ndarray, optional): The output array. If not None, the distance matrix Y is stored in this array. **kwargs (dict, optional): Extra arguments to `metric`: refer to each metric documentation for a list of all possible arguments. Some possible arguments: p (float): The p-norm to apply for Minkowski, weighted and unweighted. Default: 2.0 Returns: Y (cupy.ndarray): A :math:`m_A` by :math:`m_B` distance matrix is returned. For each :math:`i` and :math:`j`, the metric ``dist(u=XA[i], v=XB[j])`` is computed and stored in the :math:`ij` th entry. )float32float64rlrrSrTz=XA and XB must have the same layout (XA.order=%s, XB.order=%sz!XA must be a 2-dimensional array.z!XB must be a 2-dimensional array.rWzHXA and XB must have the same number of columns (i.e. feature dimension.)rra@Nz1out must have same layout as input (out.order=%s)F)copyrXzUnknown Distance Metric: %sz/2nd argument metric must be a string identifier)rRrPrOrrrkr\r]shapelenastyperesize isinstancestrlower _METRIC_ALIASgetr^r TypeError)XAXBmetricoutrXA_orderXB_orderssBmAmBra out_ordermstrrrdr r r cdist(sj        r)r~cKs,t||f||d|}t|d}||S)a[Compute distance between observations in n-dimensional space. Args: X (array_like): An :math:`m` by :math:`n` array of :math:`m` original observations in an :math:`n`-dimensional space. Inputs are converted to float type. metric (str, optional): The distance metric to use. The distance function can be 'canberra', 'chebyshev', 'cityblock', 'correlation', 'cosine', 'euclidean', 'hamming', 'hellinger', 'jensenshannon', 'kl_divergence', 'matching', 'minkowski', 'russellrao', 'sqeuclidean'. out (cupy.ndarray, optional): The output array. If not None, the distance matrix Y is stored in this array. **kwargs (dict, optional): Extra arguments to `metric`: refer to each metric documentation for a list of all possible arguments. Some possible arguments: p (float): The p-norm to apply for Minkowski, weighted and unweighted. Default: 2.0 Returns: Y (cupy.ndarray): Returns a condensed distance matrix Y. For each :math:`i` and :math:`j` and (where :math:`i < j < m`), where m is the number of original observations. The metric ``dist(u=X[i], v=X[j])`` is computed and stored in entry ``m * i + j - ((i + 2) * (i + 1)) // 2``. )r}r~rW)rrtriu_indices_from)rr}r~rall_distup_idxr r r pdists rrocCsPt|}|j\}}t|}|j\}}||kr td||ft||d|dS)aCompute the distance matrix. Returns the matrix of all pair-wise distances. Args: x (array_like): Matrix of M vectors in K dimensions. y (array_like): Matrix of N vectors in K dimensions. p (float): Which Minkowski p-norm to use (1 <= p <= infinity). Default=2.0 Returns: result (cupy.ndarray): Matrix containing the distance from every vector in `x` to every vector in `y`, (size M, N). zGx contains %d-dimensional vectors but y contains %d-dimensional vectorsr?)r}ra)rrkrqr]r)xyrarkrkkr r r distance_matrixs    r)r9N)r9)ro)$ __future__rrrOrP cuvs.distancer ImportErrorpylibraft.distancer rr _METRIC_INFOS_METRICSdictrxlistkeys_METRICS_NAMESrRr?r$r%r*r.r0r2r9r=rBrCrFrGrrrr r r r s       ;  $!### ! "X!