@iPdZddlZddlZddlmZmZmZmZm Z ddl m Z eje Z GddZGddZy) z:Iteration manager for Stage4 iterative interaction engine.N)DictAnyListOptionalTuple)datetimecFeZdZdZdededefdZdeeeffdZ de fd Z y ) IterationStatez-State object for tracking iteration progress.user_idquery max_roundscX||_||_||_d|_d|_g|_y)zInitialize iteration state. Args: user_id: User identifier query: Original user query max_rounds: Maximum number of iteration rounds rN)r r r current_round user_vectorhistory)selfr r r s S/home/ubuntu/codebase/yexijia/保研/colocation_mvp/controller/iteration_manager.py__init__zIterationState.__init__s0  $ round_datac|jj|jtjj d|y)zcAdd a round to history. Args: round_data: Data for this round )round timestampN)rappendrrnow isoformat)rrs r add_roundzIterationState.add_rounds@ ''!113   rreturnc4|j|jk\S)zrCheck if iteration is complete. Returns: True if current_round >= max_rounds )rr )rs r is_completezIterationState.is_complete*s !!T__44rN) __name__ __module__ __qualname____doc__strintrrrrboolr rrr r s?7CS  DcN  5T5rr c eZdZdZ ddefdZdededeejfdZ dd e d ee ee fde ee ffd Zd e d e eeefdejfd Z dd e d ee ee fde ee ffdZ dd e d e eeefdede ee ffdZ ddededed ee ee fde ee ff dZ dd e d ee ee fde ee ffdZy)IterationManagerz5Manager for iterative preference refinement (Stage4).N fusion_alphac t||_||_||_||_||_||_||_||_y)uInitialize iteration manager. Args: miner: CoLocationMiner instance learner: PreferenceLearner instance memory: MemoryStore instance embedder: PatternEmbedder instance llm_client: Optional LLMClient for intent understanding intent_encoder: Optional IntentEncoder for Stage0 intent_mapper: Optional IntentMapper for Stage0 fusion_alpha: Weight for LLM intent vs feedback fusion (α) N)minerlearnermemoryembedder llm_clientintent_encoder intent_mapperr+) rr-r.r/r0r1r2r3r+s rrzIterationManager.__init__6s?     $,*(rr r rcP|j |jtjdy tjd|ddd|jj |}|stj dy|j dg}|rt|dk(r KK\]`af]g\hhij kL  NNSTUSVWbfN g s-AG8?AG8BG8BG88 H%H  H%state mining_paramsc  |xjdz c_tjd|jd|j|dddd}|jj |j d d|j d d|j d d}tjd |jd t|d|dd}|D]*}d|vs|jj|d}||d<,d}|j<|jj||j}tjdnL|jj||j|jdz }|tjd|}i} |tt!||dd} | D cgc]\} } |  }} } | D]=\} } dj#t| j dg}t%| | |<?tjdt|dntjd|j dd}|jj'||}|j||t|t|| |j2t%t(j*j-|jndd}|j/t|t||d d!|Scc} } w)"aRun one iteration round. Algorithm: 1. Mine candidate patterns 2. Rank patterns by current user vector 3. Return ranked patterns for user feedback Args: state: Current iteration state mining_params: Optional mining parameters (if None, uses defaults) Returns: Dictionary containing ranked patterns and metadata zStage4: Starting round /N333333? confidencemin_participationmax_pattern_sizepriorityrVrWrXStage4: Round z mined z patterns embeddingpatternz/Stage4: Scored patterns using state user vectorr interaction_roundz4Stage4: Scored patterns using learner's fused vectorc |dSNrPr(xs rz0IterationManager.run_one_round..adrTkeyreverse,zStage4: Re-ranked z8Stage4: No user vector available, using original rankingmin_confidence?ri)rpatternsrulestotal_patterns total_rulessimilarity_scoresuser_vector_normrr)rorprr)rr;r<r r- mine_patternsr?r@r0encode_patternrr._score_with_vectorscore_patternsr sortedzipjoinfloatgenerate_rulesrDrErFr)rrMrNrm top_patternsr\ pattern_vecscoresranked_patternsrqrankedp_score pattern_keyrirn round_results r run_one_roundzIterationManager.run_one_rounds q  -e.A.A-B!EDTDTCUVW  &)$%(M ::+++//0CSI*../A1E"&&z<@,   nU%8%8$9XyYZ } $G')"mm::79;MN '2 $$     (\\44\5CTCTUF KKI K\\00 "'"5"5"91F ! RT'  L&)"F .44VTQqVO4#5!hhvaeeIr.B'CD 16u!+.# KK,S-A,B)L M KKR S'**+;SA ))/.)Y(('!(mu:!2LQL]L]LibiinnU5F5F&G Hor  !(mu: ,-? @  C5s8 Lfeedbackc|jdg}|jdg}|D]W}|jd}|jj|}|2|jj |j |Y|D]W}|jd}|jj|}|2|jj|j |Y|jj|j }|jj|j } d} | r)| jdrtj| d} | H|F|j| zd|jz |zz} tjd|jdnx||} tjd n^| | } tjd nD|jj!} tj"| } tj%d | |_tjd tj(j+| d d| S)uHUpdate user vector based on feedback. Formula: u_t = α * u_llm + (1-α) * u_feedback Args: state: Current iteration state feedback: Dictionary with 'positive' and 'negative' pattern lists Returns: Updated user preference vector positivenegativerhNrJrPz!Stage4: Fused user vector (alpha=r8z2Stage4: Using feedback vector only (no LLM intent)z2Stage4: Using LLM intent vector only (no feedback)z3Stage4: No user vector available, using zero vectorz"Stage4: Updated user vector (norm=r7)r?splitr0rtr/ add_positiver add_negativer.compute_feedback_vector load_intentrDarrayr+r;r< get_sentence_embedding_dimensionzerosr>rrErF) rrMrpositive_patternsnegative_patternsr pattern_listr} u_feedback intent_datarJu_tdims rupdate_user_vectorz#IterationManager.update_user_vectors%LLR8$LLR8-K&,,S1L--66|DK& (( D --K&,,S1L--66|DK& (( D -\\99%--H kk--emm<  ;??73HH[12E  !7##e+q43D3D/D .RRC KK;Dr%rzrDrErF) rrMrrrupdated_vectortraining_resultrrKs rupdate_and_trainz!IterationManager.update_and_trainWsU 00A    A%BUBUAVV_`a!--*!"( $-$+KK$;#  Hhohshstzh{U[]_I`acIdBGIHHIJK!/O JXIcbiinn^&D E'  il'   !:1#>N$ V# sB C D%(DDKc 8tjd|dt|||}|j|||_g}|j sZ|j ||}|j|tjd|jd|d|j sZ|j||}||||d|d|jdi|j4ttjj|jd Sd d S) aRun iterative preference refinement. Algorithm: Iterative Preference Refinement Input: Q (query), K (rounds) Output: P* (best patterns) Initialize u0 for t=1..K: Generate Ct (candidate patterns) Rank Ct by ut Collect Ft (feedback) Update ut Return best Args: query: User query user_id: User identifier K: Number of iteration rounds mining_params: Optional mining parameters Returns: Dictionary containing final results and iteration history z)Stage4: Starting iterative refinement (K=r8rYrQz completedrmrnrqrl)r total_roundsiteration_historyfinal_patterns final_rulesfinal_similarity_scoresrr)r;r<r rLrr rrr _final_rankr?rzrDrErF) rr r rrNrMrr final_results rrunzIterationManager.runs(6  ?s!DEwq1!11%A##%--e]CL  $ $\ 2 KK.)<)<(=QqcL M ##%''}= !2*:6'0'3'7'78KR'PLQL]L]LibiinnU5F5F&G H  ps  rc ntjd|dddd}|jj|j dd|j dd|j d d}|dd }|D]*}d |vs|j j |d }||d <,d}|j'|jj||j}n2|jj||j|j }|}i} |ott||dd} | D cgc]\} } |  }} } | D]=\} } djt| j d g}t!| | |<?|j dd}|jj#||}||| dScc} } w)a Generate final ranking after all iterations. Args: state: Final iteration state mining_params: Optional mining parameters Returns: Dictionary containing final ranked patterns and rules z Stage4: Generating final rankingNrRrSrTrUrVrWrXrZr[r\r]c |dSr`r(ras rrcz.IterationManager._final_rank..rdrTrerhrirjrk)rmrnrq)r;r<r-rsr?r0rtrr.rurvr rrwrxryrzr{)rrMrNrmr|r\r}r~rrqrrrrrrirns rrzIterationManager._final_ranks  67  %($%(M::+++//0CSI*../A1E"&&z<@,  } #G')"mm::79;MN '2 $$     (\\44\5CTCTUF\\00 "'"5"51F '  L&)"F .44VTQqVO4"5!hhvaeeIr.B'CD 16u!+.#'**+;SA ))/.)Y(!2  5s+ F1)NNNrR)N)NN)NrS)r!r"r#r$rzrr%rrDndarrayrLr rrrrrrr&rrrr(rrr*r*3s?DH48'*)$)0=c=C=HRZZh$,T#s(^$<hHLSRUXhT>>$(d3i$8>=?ZZ>BJN'+ N .6tCH~.F 04S#X D!%/01 n1 "&sDI~"61 *-1 6:#s(^1 h7;9 9 s9 s9 #DcN39 ?CCH~9 x>BB B "*4S>":B FJ3PS8nB rr*)r$loggingnumpyrDtypingrrrrrr getLoggerr!r;r r*r(rrrs@@33   8 $%5%5PT T r