@i nSrSSKrSSKJrJrJrJr SSKr\R"\ 5r SS\\ \4S\ S\R4SjjrS\RS \RS\4S jrSS\RS \RS \\S\4S jjrSS \\\ \4S \RS \\S\S\\4 Sjjrg)zAPreference scorer for ranking patterns based on user preferences.N)ListDictAnyOptionalmemorydimreturncURS/5nURS/5n[U5S:Xa%[U5S:Xa[R"U5$[U5S:a,[R"U5n[R "USS9nO[R"U5n[U5S:a,[R"U5n[R "USS9nO[R"U5nXW- nU$)aBuild user preference vector from memory. Formula: u = mean(positive_vectors) - mean(negative_vectors) Args: memory: User memory dictionary with 'positive' and 'negative' lists dim: Dimension of the embedding vectors Returns: User preference vector (anchor vector) positivenegativer)axis)getlennpzerosarraymean) rrr r pos_vectorsmean_pos neg_vectorsmean_neg user_vectors D/home/ubuntu/codebase/yexijia/保研/iCoLoc/src/preference/scorer.pybuild_user_vectorr szz*b)Hzz*b)H 8}c(mq0xx} 8}qhhx( 77;Q/88C= 8}qhhx( 77;Q/88C=%K  pattern_vecuser_vecc[RRU5n[RRU5nUS:a#US:a[R"X5X#-- nU$SnU$)zScore pattern using baseline method (cosine similarity). Args: pattern_vec: Pattern embedding vector user_vec: User preference vector Returns: Cosine similarity score r)rlinalgnormdot)rr pattern_norm user_norm similaritys rscore_with_baseliner&0sa99>>+.L x(IaIMVVK2l6NO   rmodelc|Uc [X5$SSKnURU5RS5nURU5RS5nUR 5 U"U5nU"U5nSSS5 UR WSS9nUR WSS9n US:aKU S:aEUR UR5UR55X-- n U R5n U $Sn U $!,(df  N=f![a.n [RSU S35 [X5sSn A $Sn A ff=f)aScore pattern using learned preference encoder model. Args: pattern_vec: Pattern embedding vector user_vec: User preference vector model: Learned preference encoder model (PyTorch nn.Module) Returns: Cosine similarity score after projection Nr)rrzError in model-based scoring: z, falling back to baseline) r&torch FloatTensor unsqueezeno_gradr!r"squeezeitem Exceptionloggerwarning) rrr'r*pattern_tensor user_tensor pattern_proj user_projpattern_proj_normuser_proj_normr%es rscore_with_modelr:Es9 }";99:**;7AA!D''1;;A> ]]_ 0Lk*I "JJ|J;I15 q ^a%7<#7#7#99;L;L;NO!2J$*JJ!_$ :7s:TUV";99:s=AD$C25A8D.D2 D<D D; #D60D;6D;patternsalphac/nUHenURS5nUcSnOUb [XaU5nO [Xa5nURSS5nX7-SU- U--n URU 5 Mg U$)abScore patterns with preference and confidence fusion. Final score = alpha * preference_score + (1 - alpha) * confidence Args: patterns: List of pattern dictionaries user_vec: User preference vector model: Optional learned preference encoder model alpha: Weight for preference score (default: 0.7) Returns: List of final scores embeddingr confidencer))rr:r&append) r;rr'r<scorespatternrpreference_scorer? final_scores rscore_patternsrEts"Fkk+.  "  #3K5#Q $7{#M [[s3 .!e)z1II  k"+. Mr)i)N)Ngffffff?)__doc__numpyrtypingrrrrlogging getLogger__name__r1strintndarrayrfloatr&r:rErrrQsG,,   8 $#d38n#3##LRZZ2::%,-1,:"**,: ,:$SM,:5:,:b*.!$*T$sCx.1*JJ*!#***.e*r