o pi@sjddlZddlZddlZdejfddZdddZd d Zdd d Z d dZ dddZ ddZ ddZ dS)NXcCs0ttj|ddd}d||dk<|j|jS)zL2 normalize vectors Parameters ---------- X : `np.ndarray` (n_samples, n_dimensions) vectors. Returns ------- normalized : `np.ndarray` (n_samples, n_dimensions) L2-normalized vectors )axis?r)npsqrtsumT)rnormr P/home/ubuntu/.local/lib/python3.10/site-packages/pyannote/core/utils/distance.py l2_normalize"s  r euclideanFcCsf|dkr |rdSdtjfS|dkr|rdSdtjfS|dkr dS|dkr)dtjfSd|d }t|) abReturn range of possible distance between two vectors Parameters ---------- metric : `str`, optional Metric. Defaults to 'euclidean' normalize : `bool`, optional Set to True if vectors are L2-normalized. Defaults to False. Returns ------- min_dist, max_dist : `float` Range of possible distance. r)g@r sqeuclidean)rg@cosineangularzdist_range does not support z metric.)rinfpiNotImplementedError)metric normalizemsgr r r dist_range6s    rcCs^|j\}|dkr tgSg}t|dD]}|||||dd}||qt|S)zHelper function for pdistrrN)shaperarrayrangeappendhstack)rfuncn_items distancesidistancer r r _pdist_func_1DZs   r%cKs|dkrtjjj|fddi|}ttd|ddS|dkr5|jdks.Jd|d t|d d S|d krL|jdksFJd|d t|tj S|d krc|jdks]Jd|d t|tj S|dkr{|jdkstJd|d t|dd Stjjj|fd|i|S)arSame as scipy.spatial.distance with support for additional metrics * 'angular': pairwise angular distance * 'equal': pairwise equality check (only for 1-dimensional fX) * 'minimum': pairwise minimum (only for 1-dimensional fX) * 'maximum': pairwise maximum (only for 1-dimensional fX) * 'average': pairwise average (only for 1-dimensional fX) rrrrequalr'z(' metric only supports 1-dimensional fX.cS||kSNr xrr r r {zpdist..minimummaximumaveragecS d||SN?r r+r r r r- ) scipyspatialr$pdistrarccosclipndimr%r/r0)fXrkwargsrr r r r8ks    r8cstfddt|DS)zHelper function for cdistc3s|]}|VqdSr*r ).0x_trnX_tstr r r sz!_cdist_func_1D..)rvstackiter)X_trnrAr r r@r _cdist_func_1DsrFcKsD|dkrtjjj||fddi|}ttd|ddS|dkr<|jdkr,|jdks4Jd|d t||d d S|d krY|jdkrJ|jdksRJd|d t||tj S|d krv|jdkrg|jdksoJd|d t||tj S|dkr|jdkr|jdksJd|d t||dd Stjjj||fd|i|S)axSame as scipy.spatial.distance.cdist with support for additional metrics * 'angular': pairwise angular distance * 'equal': pairwise equality check (only for 1-dimensional fX) * 'minimum': pairwise minimum (only for 1-dimensional fX) * 'maximum': pairwise maximum (only for 1-dimensional fX) * 'average': pairwise average (only for 1-dimensional fX) rrrrr&r'rr(z7' metric only supports 1-dimensional fX_trn and fX_tst.cSr)r*r r?rAr r r r-r.zcdist..r/r0r1cSr2r3r rGr r r r-r5) r6r7r$cdistrr9r:r;rFr/r0)fX_trnfX_tstrr=rr r r rHs4      rHcCstt|t|}}t||krtdt||t||}}t||||dd|d|dS)aCompute index in condensed pdist matrix V 0 | . 0 1 2 3 -> 1 | . . 4 5 6 <- ==> 0 1 2 3 4 5 6 7 8 9 2 | . . . 7 8 ^ 3 | . . . . 9 4 | . . . . . ---------- 0 1 2 3 4 Parameters ---------- n : int Number of inputs in squared pdist matrix i, j : `int` or `numpy.ndarray` Indices in squared pdist matrix Returns ------- k : `int` or `numpy.ndarray` Index in condensed pdist matrix zi and j should be different.rr)rrany ValueErrorr/r0int64)nr#jr r r to_condenseds .rQc Csxt|}t|td|d|dd|ddd}t|dd||d|d|d}||fS)a1Compute indices in squared matrix Parameters ---------- n : int Number of inputs in squared pdist matrix k : `int` or `numpy.ndarray` Index in condensed pdist matrix Returns ------- i, j : `int` or `numpy.ndarray` Indices in squared pdist matrix irrr4rK)rrrNr)rOkr#rPr r r to_squareds 8.rT)rF)r)numpyrscipy.spatial.distancer6scipy.cluster.hierarchyndarrayrrr%r8rFrHrQrTr r r r s $ " ,