o {i} @sddlmZddlZddlZddlmZmZddlm Z m Z m Z e ej jZe ej jZejdddd d Zejddd d Zejddd dZejddddZejdejejjejjddddejjejjddddgejejddddZejejejejjddejejj ddfejjejjddddejjejj ddddejjejjddddejjejj ddddgdejjdddejj dddejj ejjejjejjejjdddddZejddddZejejejejjejejj fejjejjddddejjejj ddddejjejjddddejjejj ddddgdejj ejjejjejjejjdddd d!Z ejej ejjejjddddejjejj ddddejjejjddddejjejj ddddgdejj ejj ejjejjejjejjd"ddd#d$Z!ed%d&Z"ejdd'd(d)Z#ejd*ej ejjejjddddejjejj ddddejjejjddddejjejj ddddgdejj dddejj ejj ejj$ejjd+d,d-d.Z%ed/d0Z&ed1d2Z'edd4d5Z(ed6d7Z)ed8d9Z*ejd*ej ejjejjddddejjejj ddddejjejjddddejjejj ddddgdd'd:d;Z+edd,d?d@Z-ej.dd'dAdBZ/edCdDZ0edEdFZ1edGdHZ2edIdJZ3edKdLZ4edMdNZ5edOdPZ6edQdRZ7ejdejj ejj ejj ejj$ejjdSd,dTdUZ8ej.dddVdWdXZ9edYdZZ:ejdd[ejj id,d\d]Z;ed^d_Zej.dddVdedfZ?edgdhZ@didjZAee@fdkdlZBeddmdnZCedodpZDedqdrZEejddsdt uddvdwZFejddsdt uddxdyZGidze#d{e#d|e%d}e&d~e&de&de'de'de'de'de(de*de+de)de,de1de0ide2de3de4de5de6de7de<deBdeBdeCdeCdeCdeCde=deDdeDdeEeEeEdZHdZIe%ejJde%ejJde8e9de;e9de>e?de-e/ddZKdS))print_functionN)normtau_rand) kantorovichjensen_shannon_divergencesymmetric_kl_divergenceT)cacheh㈵>:0yE>cCs$t||}|||t|kSN)npabs)abrtolatoldiffrF/home/ubuntu/.local/lib/python3.10/site-packages/pynndescent/sparse.pyisclosesrcCs@t|}ttjdtjd|dd|ddkf}||S)Ndtype)r sort concatenateonesbool_)arrauxflagrrr arr_uniques .r!cCs6|jddkr |S|jddkr|Stt||fSNr)shaper!r r)ar1ar2rrr arr_union$s r&cCs:t||f}||dd|dd|ddkS)Nrr)r rr)r$r%rrrr arr_intersect0s$r'zi4(i4[:],i4[:])rC)readonly)i1i2)localsc Cs|jddks|jddkrdSd}d}|jdd}|jdd}||}||}d} ||krU|d7}||krB|d7}||}n |S||krR|d7}||}n( |S||krf||krf|d7}||}n||krw||krw|d7}||}n |Sq-Nrrr#) r$r%r*r+limit1limit2j1j2resultrrrfast_intersection_size8s>     r4) result_ind result_datavalr*r+r1r2)fastmathr,rc Cs|jd|jd}tj|tjd}tj|tjd}d}d}d} ||jdkr||jdkr||} ||} | | kr[||||} | dkrR| || <| || <| d7} |d7}|d7}n5| | krx||} | dkrs| || <| || <| d7} |d7}n||} | dkr| || <| || <| d7} |d7}||jdkr||jdks.||jdkr||} ||} | dkr| || <| || <| d7} |d7}||jdks||jdkr||} ||} | dkr| || <| || <| d7} |d7}||jdks|d| }|d| }||fSNrrr)r#r zerosint32float32) ind1data1ind2data2 result_sizer5r6r*r+nnzr1r2r7rrr sparse_sumnsh    rCcCst|||| Sr )rC)r=r>r?r@rrr sparse_diffrD)r7r*r+r1r2c Cstjjtjj}tjjtjj}d}d}||jdkrj||jdkrj||}||} || krO||||} | dkrF|||| |d7}|d7}n || krX|d7}n|d7}||jdkrj||jdks$||fSr-) numbatypedList empty_listtypesr;r<r#append) r=r>r?r@r5r6r*r+r1r2r7rrr sparse_muls&    rL)r3r7r*r+r1r2c Cs|jd}|jd}d}d}d}||} ||} | | krF||||} || 7}|d7}||kr3|S||} |d7}||krA|S||} n!| | krY|d7}||krT|S||} n|d7}||krc|S||} q)NrTrr.) r=r>r?r@dim1dim2r3r*r+r1r2r7rrrsparse_dot_products:    rPc Cst||}tj|jdtjd}tj|jdtjd}d}d}d} ||jdkr||jdkr||} ||} | | kr`||||} | dkrW|||| <|||| <| d7} |d7}|d7}n1| | kr{||} | dkrv|||| <| d7} |d7}n||} | dkr|||| <| d7} |d7}||jdkr||jdks/||jdkr||} | dkr|||| <| d7} |d7}||jdks||jdkr||} | dkr|||| <| d7} |d7}||jdks|d| }|d| }||fSr9)r&r r:r#r<) r=r>r?r@r5 result_data1 result_data2r*r+rBr1r2r7rrr dense_union<s\           rS)r8cCsDt||||\}}d}t|jdD] }|||d7}qt|S)NrMr)rDranger#r sqrtr=r>r?r@_aux_datar3irrrsparse_euclideanvs  r[z#f4(i4[::1],f4[::1],i4[::1],f4[::1]))rYr3rdimrZ)r8r,c CsDt||||\}}d}t|}t|D] }|||||7}q|SNrM)rDlenrU) r=r>r?r@rXrYr3r\rZrrrsparse_squared_euclideans  r_cCs@t||||\}}d}t|jdD] }|t||7}q|SNrMrrDrUr#r r rWrrrsparse_manhattans rbcCsBt||||\}}d}t|jdD] }t|t||}q|Sr`)rDrUr#maxr r rWrrrsparse_chebyshevs rd@c CsLt||||\}}d}t|jdD] }|t|||7}q|d|S)NrMr?ra) r=r>r?r@prXrYr3rZrrrsparse_minkowskis  rhcCs$t||||djd}t||Sr")rDr#float)r=r>r?r@ n_features num_not_equalrrrsparse_hammings rlcCs~t|}t|}t||||\}}d|tj}t||||\}} t| } t|| ||\} } d} | D]} | | 7} q6| S)NrfrM)r r rCastyper<rDrL)r=r>r?r@ abs_data1 abs_data2 denom_inds denom_data numer_inds numer_datarXval_datar3r7rrrsparse_canberras    ruc Csvt||||\}}t|}|jddkrdSt|}|dkr"dSt||||\}}t|}t|}t||SNrrM)rCr r r#sumrDri) r=r>r?r@rXrq denominatorrs numeratorrrrsparse_bray_curtiss     rzcCs>t||}|jd|jd|}|dkrdSt|||Srvr4r#rir=r>r?r@ num_equal num_non_zerorrrsparse_jaccards r)r~r}cCsJt||}|jd|jd|}|dkrdS|dkrtSt|| Srv)r4r# FLOAT32_MAXr log2r|rrrsparse_alternative_jaccards rcCsdtd| S)Nrfre)pow)vrrrcorrect_alternative_jaccardrErcCs6t||}|jd|jd|}||}t||Sr"r{r=r>r?r@rj num_true_truer~rkrrrsparse_matchings  rcCFt||}|jd|jd|}||}|dkrdS|d||S)NrrMrer4r#r=r>r?r@rr~rkrrr sparse_dice! rcCsNt||}|jd|jd|}||}|dkrdSt|||||Srvr{rrrrsparse_kulsinski-s rcC:t||}|jd|jd|}||}d|||SNrrerrrrrsparse_rogers_tanimoto; rcCsh|jd|jdkrt||krdSt||}|t|dkkr,|t|dkkr,dSt|||Srv)r#r allr4rwri)r=r>r?r@rjrrrrsparse_russellraoDs " $rcCrrrrrrrsparse_sokal_michenerQrrcCr)NrrMg?rrrrrsparse_sokal_sneathZrrc Cspt||||\}}d}t|}t|}|D]} || 7}q|dkr&|dkr&dS|dks.|dkr0dSd|||SNrMrf)rLr) r=r>r?r@rXrYr3norm1norm2r7rrr sparse_cosinefs r)r3norm_xnorm_yr\rZc Cst||||\}}d}t|}t|}t|} t| D]} ||| 7}q|dkr.|dkr.dS|dks6|dkr8tS|dkr>tS|||}t|Sr])rLrr^rUrr r) r=r>r?r@rXrYr3rrr\rZrrrsparse_alternative_cosinexs    r)r8rcCs.tdt|dds |dkrdSdtd| SNrMgHz>)rrfre)rr rdrrr!sparse_correct_alternative_cosinesrcCst||||}d|S)Nrf)rPr=r>r?r@r3rrr sparse_dotsrr3cCs&t||||}|dkr tSt| Sr])rPrr rrrrrsparse_alternative_dots rcCsFd}d}d}|jddkr|jddkrdS|jddks$|jddkr&dSt|jdD]}|||7}q-t|jdD]}|||7}q=||}||}tj|jdtjd} tj|jdtjd} t|jdD] }|||| |<qkt|jdD] }|||| |<q}tt| d||jd|d} tt| d||jd|d} t|| || \} }t| }|D]}||7}qt|jdD]}|||vr|| ||8}qt|jdD]}|||vr|| ||8}qt ||}|||||jd7}| dkr| dkrdS|dkrdSd|| | S)NrMrrfrrT) r#rUr emptyr<rVrrLsetr&)r=r>r?r@rjmu_xmu_y dot_productrZ shifted_data1 shifted_data2rr dot_prod_inds dot_prod_datacommon_indicesr7 all_indicesrrrsparse_correlationsX       rc Cst||||\}}d}t|}t|}t||} |D] } |t| 7}q|dkr2|dkr2dS|dks:|dkrr?r@aux_indsrYr3rrsqrt_norm_prodr7rrrsparse_hellingers  r)r3 l1_norm_x l1_norm_yr\rZc Cst||||\}}d}t|}t|}t|} t| D] } |t|| 7}q|dkr3|dkr3dS|dks;|dkr=tS|dkrCtSt|||}t|Sr`)rLr rwr^rUrVrr) r=r>r?r@rrYr3rrr\rZrrrsparse_alternative_hellingers     rcCs4tdt|dds |dkrdStdtd| Sr)rr r rVrrrrr$sparse_correct_alternative_hellinger'srcCst||k Sr )r r<)xyrrrdummy_ground_metric/rErcstfdd}|S)aGenerate a "ground_metric" suitable for passing to a ``sparse_kantorovich`` distance function. This should be a metric that, given indices of the data, should produce the ground distance between the corresponding vectors. This allows the construction of a cost_matrix or ground_distance_matrix between sparse samples on the fly -- without having to compute an all pairs distance. This is particularly useful for things like word-mover-distance. For example, to create a suitable ground_metric for word-mover distance one would use: ``wmd_ground_metric = create_ground_metric(word_vectors, cosine)`` Parameters ---------- ground_vectors: array of shape (n_features, d) The set of vectors between which ground_distances are measured. That is, there should be a vector for each feature of the space one wishes to compute Kantorovich distance over. metric: callable (numba jitted) The underlying metric used to cpmpute distances between feature vectors. Returns ------- ground_metric: callable (numba jitted) A ground metric suitable for passing to ``sparse_kantorovich``. cs||Sr r)index1index2ground_vectorsmetricrr ground_metricQsz+create_ground_metric..ground_metric)rFnjit)rrrrrrcreate_ground_metric4srcCsht|jd|jdf}t|jdD]}t|jdD]}||||||||f<qqt|||Sr")r rr#rUr)r=r>r?r@r cost_matrixrZjrrrsparse_kantorovichXs  rcCs(d}d}d}d}d} d} d} t|} t|} dd}||jdkr| |jdkr||}|| }||kr`||||7}| ||| 7} | || | 7} || | |}|}|d7}| d7} n?||kr||||7}| ||| 7} || | |}|}|d7}n||||7}| || | 7} || | |}|}| d7} ||jdkr| |jdks*||jdkr||}||||7}| ||| 7} || | |}|}|d7}||jdks| |jdkr || }||||7}| || | 7} || | |}|}| d7} | |jdkst|d|S)NrMrcSstt||Sr )r powerr )rrgrrrosz'sparse_wasserstein_1d..rrf)r rwr#r)r=r>r?r@rgr3old_inddeltar*r+cdf1cdf2rrrr1r2rrrsparse_wasserstein_1dcsd    rcCt||||\}}t||Sr )rSrr=r>r?r@ dense_data1 dense_data2rrr sparse_jensen_shannon_divergence rcCrr )rSrrrrrsparse_symmetric_kl_divergencerrF)parallelrrfc Cst|jdD]}||dfg} ||dfg} td|jdD]} ||| fdkr,nd} tt| D]e} | | }||||| f|||| fd}||||| f|||| fd}|||||d}|||||d}|||||}| | tkr|||| fkrt||krd} nq4| r| ||| f| ||| fq t|jdD]&} | t| kr| | ||| f<| | ||| f<qd||| f<tj ||| f<qq||fS)NrrTFr) rFpranger#rUr^ FLOAT32_EPSrrKr inf)indices distances data_indices data_indptr data_datadist rng_stateprune_probabilityrZ new_indices new_distancesrr kcfrom_ind from_datato_indto_datarrrr diversifysF ""   rc Cs|jdd} t| D]} ||| || d} ||| || d} t| } tj| jdtjd}td| jdD]n}| |}t|D]c}| |}||dkr| |}| |}|||||d}|||||d}|||||d}|||||d}|||||}| |tkr|| |krt ||krd||<nqHq>t| jdD]}| |}||dkrd||| |<qq dS)Nrrr) r#rFrr argsortrint8rUrr) graph_indptr graph_indices graph_datarrrrrrn_nodesrZcurrent_indices current_dataorderretainedidxrrlrgq from_indsrto_indsrrrrr diversify_csrs<      r euclideanl2 sqeuclidean manhattanl1taxicab chebyshevlinflinfty linfinity minkowskicanberra braycurtishammingjaccarddicematching kulsinskirogerstanimoto russellrao sokalmichener sokalsneathcosine correlationr wassersteinwasserstein_1dzwasserstein-1dzkantorovich-1d hellingerzjensen-shannonjensen_shannonz symmetric-kl) symmetric_klsymmetric_kullback_liebler)rr r r r r r)r correction)rrrdotrr)r r )re)r)rf)L __future__rnumpyr rFpynndescent.utilsrrpynndescent.distancesrrrfinfor<epsrrcrrrr!r&r'rJr;Arrayuint16r4TuplerCrDListTyperLrPrSr[intpr_rbrdrhrlrurzrr vectorizerrrrrrrrrrrrrrrrrrrrrrrrrsparse_named_distancessparse_need_n_featuresrV!sparse_fast_distance_alternativesrrrrsp         )    ?    '  9                           <    $ ?    6 6     !"#$%&'(,