o pip6@sddlZddlZddlmZddlmZmZmZmZm Z m Z m Z ddl Z ddlZddlZddlZddlmZddlmZddlmZmZddlmZmZddlmZmZmZdd lmZddlmZmZmZdd lm Z d d l!m"Z"d d lm#Z#ej$%ej$j&GdddZ'dS)N)Path)IterableOptionalCallable GeneratorMappingUnionDict)ExperimentalWarning) BasePruner) BaseSampler TPESampler)Trial FixedTrial) RDBStorageJournalStorageJournalFileStorage)tqdm) bayes_mvs)Pipeline) PipelineInputc@s6eZdZdZ      d dedeedeedeeee fdeeee fd ee d e fd d Z ed efddZed efddZed efddZ d!deedee efd eegeffddZ   d"deede dedee efd ef ddZ  d#deededee efd eeddffddZdS)$ OptimizeracPipeline optimizer Parameters ---------- pipeline : `Pipeline` Pipeline. db : `Path`, optional Path to trial database on disk. Use ".sqlite" extension for SQLite backend, and ".journal" for Journal backend (prefered for parallel optimization). study_name : `str`, optional Name of study. In case it already exists, study will continue from there. # TODO -- generate this automatically sampler : `str` or sampler instance, optional Algorithm for value suggestion. Must be one of "RandomSampler" or "TPESampler", or a sampler instance. Defaults to "TPESampler". pruner : `str` or pruner instance, optional Algorithm for early pruning of trials. Must be one of "MedianPruner" or "SuccessiveHalvingPruner", or a pruner instance. Defaults to no pruning. seed : `int`, optional Seed value for the random number generator of the sampler. Defaults to no seed. average_case : `bool`, optional Optimize for average case (default). Set to False to optimize for worst case. NTpipelinedb study_namesamplerprunerseed average_casec Cs||_||_|durd|_n4t|jj}|dkr'tdtd|j|_n|dkr5td|j|_n |dkrBtt |j|_||_ t |t rN||_ n.t |trrz ttj||d|_ Wntyq} zd} t| d} ~ ww|dur|t|d|_ t |tr||_n%t |trz ttj||_Wnty} zd} t| d} ~ wwd|_tj|j d |j|j |j|jd |_||_dS) Nz.dbzHStorage with '.db' extension has been deprecated. Use '.sqlite' instead.z sqlite:///z.sqlitez.journal)rz8`sampler` must be one of "RandomSampler" or "TPESampler"zC`pruner` must be one of "MedianPruner" or "SuccessiveHalvingPruner"T)rload_if_existsstoragerr direction)rrstorage_rsuffixwarningswarnrrrr isinstancer rstrgetattroptunasamplersAttributeError ValueErrorr r rpruners create_study get_directionstudy_r) selfrrrrrrr extensionemsgr6O/home/ubuntu/.local/lib/python3.10/site-packages/pyannote/pipeline/optimizer.py__init__Ss^        zOptimizer.__init__returncCsDz|jj}W|Sty!|jdkrdnd}|tj}Y|Sw)zReturn best loss so farminimizer)r1 best_value Exceptionrr0npinf)r2r<r"r6r6r7 best_losss   zOptimizer.best_losscCst|jj}|jj|dS)zReturn best parameters so fartrial)rr1 best_paramsr parameters)r2rBr6r6r7rCs zOptimizer.best_paramscCs|j|jS)z8Return pipeline instantiated with best parameters so far)r instantiaterC)r2r6r6r7 best_pipelineszOptimizer.best_pipelineFinputs show_progresscsBtt}dkrdddddtdtffdd }|S) a Create objective function used by optuna Parameters ---------- inputs : `iterable` List of inputs to process. show_progress : bool or dict Show within-trial progress bar using tqdm progress bar. Can also be a **kwarg dict passed to tqdm. Returns ------- objective : `callable` Callable that takes trial as input and returns correspond loss. Tz Current trialFr)descleavepositionrBr9c sRzj}Wnty}z d}g}WYd}~nd}~wwg}g}jjj|d}dkrAtddti}|dtD]\}} t } t | t rY| di} ni} || fi| } t } | | | t }|dur|| | }| |nddlm}|| d| || d }t }| ||dkr|d jdurqE||durt|nt|||rtqEdkr||d t||d t||durtt|d kr|d}}}nt|d d\\}\}}}}n |jd d\}\}}jr|dur|St|Sj dkr'|S|S)zCompute objective value Parameter --------- trial : `Trial` Current trial Returns ------- loss : `float` Loss NrAFtotalrpipeline_kwargs) get_annotated annotation)uemrprocessing_timeevaluation_timeg?)alphar:r6)!r get_metricNotImplementedErrorrErDrlenupdate enumeratetimer'rgetappendlosspyannote.databaserNrreportr>meanabs should_pruner* TrialPrunedclose set_user_attrsumuniquerconfidence_intervalrr0)rBmetricr4lossesrQrRr progress_bariinputbefore_processingrMoutputafter_processingbefore_evaluationr\rN_after_evaluationr_ lower_bound upper_boundrGr2rHr6r7 objectivesr       "  z*Optimizer.get_objective..objective)listrVrfloat)r2rGrHn_inputsrvr6rur7 get_objectives  jzOptimizer.get_objective n_iterations warm_startcCsd|j_|j||d}|r4|j|}ttjdtd|j |Wdn1s/wY|jj ||dddd|j_|j |j d S) aTune pipeline Parameters ---------- inputs : iterable List of inputs processed by the pipeline at each iteration. n_iterations : int, optional Number of iterations. Defaults to 10. warm_start : dict, optional Nested dictionary of initial parameters used to bootstrap tuning. Returns ------- result : dict ['loss'] ['params'] nested dictionary of optimal parameters TrHignorecategoryNrn_trialstimeoutn_jobsFr\params) rtrainingrz_flattenr%catch_warningsfilterwarningsr r1 enqueue_trialoptimizer@rC)r2rGr|r}rHrvflattened_paramsr6r6r7tune2s  zOptimizer.tunec cs|j||d}z|j}Wnty"}z tj}WYd}~nd}~ww|rL|j|}ttj dt d|j |Wdn1sGwY d|j_ |j j|ddddz|j}|j}Wntyu}zWYd}~qLd}~wwd|j_ ||d VqM) a Parameters ---------- inputs : iterable List of inputs processed by the pipeline at each iteration. warm_start : dict, optional Nested dictionary of initial parameters used to bootstrap tuning. Yields ------ result : dict ['loss'] ['params'] nested dictionary of optimal parameters r~NrrTrrFr)rzr@r-r>r?rrr%rrr r1rrrrC) r2rGr}rHrvr@r4rrCr6r6r7 tune_iter^s6      zOptimizer.tune_iter)NNNNNT)F)r{NT)NT)__name__ __module__ __qualname____doc__rrrr(rr r intboolr8propertyrxr@dictrCrFrrr rrrzrrrr6r6r6r7r6s~ A       /  r)(rYr%pathlibrtypingrrrrrrr numpyr>optuna.loggingr*optuna.prunersoptuna.samplersoptuna.exceptionsr r r r optuna.trialrroptuna.storagesrrrr scipy.statsrrrrlogging set_verbosityWARNINGrr6r6r6r7s( $