QSrSSKrSSKrSSKrSSKrSSKrSSKrSSKrSSKrSSK r SSK r SSK r SSK J r SSKrSSKJr \r\ R$S:a \R&rSrSrS rS rS rS rS rSrSrSrSrSrSr Sr!Sr"\ \!4r#Sr$Sr%Sr&Sr'Sr(Sr)Sr*Sr+Sr,Sr-S r.S!r/\R`"S"S#5r1S$r2Sr&S%r3"S&S'\45r5S(r6S)r7"S*S+\85r9"S,S-\:5r;"S.S/\ 5rSlib/third_party/ml_sdk/cloud/ml/prediction/prediction_utils.py error_codePredictionError.error_codeos 99Q<  c4URSR$r!)r"rr#s r% error_messagePredictionError.error_messagess 99Q<  r(c URS$)Nr)r"r#s r% error_detailPredictionError.error_detailws 99Q<r(cNSURURUR4-$)Nz%s: %s (Error code: %d))r*r-r&r#s r%__str__PredictionError.__str__{s. %););)-):):DOO)M MNr(N)__name__ __module__ __qualname____firstlineno____doc__rFAILED_TO_LOAD_MODELINVALID_INPUTSFAILED_TO_RUN_MODELINVALID_OUTPUTSINVALID_USER_CODE FAILED_TO_ACCESS_METADATA_SERVERpropertyr&r*r-r0__static_attributes__r2r(r%rr\s:-$1.&'7a@.+0q:':D/)<1F%8F &"         Nr(ri@Bic^\rSrSrSrS SjrSrSr\S5r \S5r \S 5r S r g) Timera0Context manager for timing code blocks. The object is intended to be used solely as a context manager and not as a general purpose object. The timer starts when __enter__ is invoked on the context manager and stopped when __exit__ is invoked. After __exit__ is called, the duration properties report the amount of time between __enter__ and __exit__ and thus do not change. However, if any of the duration properties are called between the call to __enter__ and __exit__, then they will return the "live" value of the timer. If the same Timer object is re-used in multiple with statements, the values reported will reflect the latest call. Do not use the same Timer object in nested with blocks with the same Timer context manager. Example usage: with Timer() as timer: foo() print(timer.duration_secs) Nc\SUlSUlU=(d [RUlgN)startendtimeit default_timer _get_time)r$timer_fns r%__init__Timer.__init__s"DJDH5!5!5DNr(c>SUlUR5UlU$rD)rFrIrEr#s r% __enter__Timer.__enter__sDH!DJ Kr(c.UR5Ulg)NF)rIrF)r$exc_typevalue tracebacks r%__exit__Timer.__exit__s~~DH r(cxUR5nUR=(d UUR=(d U- $rDrIrFrE)r$nows r%seconds Timer.secondss* .. C HHO 1c 22r(c:[[UR-5$rD)intMICROrYr#s r% microsecondsTimer.microseconds ut||# $$r(c:[[UR-5$rD)r\MILLIrYr#s r% millisecondsTimer.millisecondsr`r(rWrD) r3r4r5r6r7rKrNrTr>rYr^rcr?r2r(r%rArAsR.6   3 3 % % % %r(rAc@\rSrSrSr\R SSj5rSrg)StatsaAn object for tracking stats. This class is dict-like, so stats are accessed/stored like so: stats = Stats() stats["count"] = 1 stats["foo"] = "bar" This class also facilitates collecting timing information via the context manager obtained using the "time" method. Reported timings are in microseconds. Example usage: with stats.time("foo_time"): foo() print(stats["foo_time"]) Nc#z# [U5nUv SSS5 WRX'g!,(df  N=f7frD)rAr^)r$namerJtimers r%time Stats.times, xE k ##DJ s ;*; 8;r2rD) r3r4r5r6r7 contextlibcontextmanagerrkr?r2r(r%rfrfs!&$$r(rfcL\rSrSrSrSrS SjrS SjrS SjrSr S S jr S r g) BaseModelz3The base definition of an internal Model interface.cXlSUlg)zdConstructs a BaseModel. Args: client: An instance of PredictionClient for performing prediction. N_client_user_processor)r$clients r%rKBaseModel.__init__s LDr(Nc g)a\Runs the preprocessing function on the instances. Args: instances: list of instances as provided to the predict() method. stats: Stats object for recording timing information. **kwargs: Additional keyword arguments for preprocessing. Returns: A new list of preprocessed instances. Each instance is as described in the predict() method. Nr2)r$rstatskwargss r% preprocessBaseModel.preprocesss r(c g)aRuns the postprocessing function on the instances. Args: predicted_output: list of instances returned by the predict() method on preprocessed instances. original_input: List of instances, before any pre-processing was applied. stats: Stats object for recording timing information. **kwargs: Additional keyword arguments for postprocessing. Returns: A new list of postprocessed instances. Nr2)r$predicted_outputoriginal_inputryrzs r% postprocessBaseModel.postprocesss r(c U=(d [5nURU5 UR[5 UR"U4SU0UD6nSSS5 UR[ 5 UR R"W4SU0UD6nSSS5 UR[5 UR"W4XS.UD6nSSS5 U$!,(df  N=f!,(df  NT=f!,(df  W$=f)z=Runs preprocessing, predict, and postprocessing on the input.ryN)rry) rf_validate_kwargsrkPREPROCESS_TIMEr{ENGINE_RUN_TIMErtpredictPOSTPROCESS_TIMEr)r$rryrz preprocessedpredicted_outputs postprocesseds r%rBaseModel.predicts  UWE&! O $__YFeFvFl % O $,,.. /#/'-/ % $ %&& N,5NFLNm &  % $ $ $ & % s#C - C*C, C C), C;cg)axValidates and sets defaults for extra predict keyword arguments. Modifies the keyword args dictionary in-place. Keyword args will be included into pre/post-processing and the client predict method. Can raise Exception to error out of request on bad keyword args. If no additional args are required, pass. Args: kwargs: Dictionary (str->str) of keyword arguments to check. Nr2)r$rzs r%rBaseModel._validate_kwargs s r(cg)aGets model signature of inputs and outputs. Currently only used for Tensorflow model. May be extended for use with XGBoost and Sklearn in the future. Args: signature_name: str of name of signature Returns: (str, SignatureDef): signature key, SignatureDef NNr2)r$rs r% get_signatureBaseModel.get_signatures r(rsrDr) r3r4r5r6r7rKr{rrrrr?r2r(r%rprps#;       r(rpcU[:H=(a9 [U[5(+=(d [UR U5S5$)aDetermines if base64 decoding is required. Returns False if framework is not TF. Returns True if framework is TF and is a user model. Returns True if framework is TF and model contains a str input. Returns False if framework is TF and model does not contain str input. Args: framework: ML framework of prediction app model: model object signature_name: str of name of signature Returns: bool r)TENSORFLOW_FRAMEWORK_NAME isinstancerpdoes_signature_contain_strr) frameworkr rs r%should_base64_decoder)sD" 0 0 O%+ + N %e&9&9.&I!&L MPr(cz[U[5(aUVs/sHn[U5PM sn$[U[5(ai[R "U5S1:Xa[ R"US5$[R"U5VVs0sHup#U[U5_M snn$U$s snfs snnf)Nb64) rlist decode_base64dictsixviewkeysbase64 b64decode iteritems)datavalkvs r%rr?sd*. /$3M# $ //$ ||DeW$   d5k **.1mmD.A B.Adaaq!!.A BB K 0 Cs B2B7c^Ucg[SURR555$)zReturn true if input signature contains a string dtype. This is used to determine if we should proceed with base64 decoding. Args: signature: SignatureDef protocol buffer Returns: bool Tc3p# UH,nUR[RR:Hv M. g7frD)dtyperstringas_datatype_enum).0rs r% -does_signature_contain_str..\s* 1/WW 66 6/s46)anyrvalues) signatures r%rrKs6   1&&--/ 1 11r(c[R"5n[R"SX5 [RR U5(d[R "U5 [RRUS5n[R"SSSX/[RS9 [R"SUU[R"5U- 5 g ![Ra [R"S5 ef=f) aCopy files from gcs to a local path. Copies files directly to the dest_path. Sample behavior: dir1/ file1 file2 dir2/ file3 copy_model_to_local("dir1", "/tmp") After copy: tmp/ file1 file2 dir2/ file3 Args: gcs_path: Source GCS path that we're copying from. dest_path: Destination local path that we're copying to. Raises: Exception: If gsutil is not found. z$Starting to copy files from %s to %s*gsutilcpz-R)stdinz"Could not copy model using gsutil.z+Files copied from %s to %s: took %f secondsN) rkloggingdebugospathexistsmakedirsjoin subprocess check_callPIPECalledProcessError exception)gcs_path dest_pathcopy_start_times r%copy_model_to_localr`s4IIK/ --6L  " "KK  WW\\(C (( $h3:D//K  --=x499;8:  & &  :;  s 'C,D cURS5(a[U[5 [n[RR U[ 5n[RR U[5n[RRU5(a2UnSSK J n [R "SU5 UR#U5$[RRU5(aSUn[R "SU5 [-US 5n [.R0"U R355sSSS5 $g![aSnSSK nSnAN![a7nSn[R"U5 [[RU5eSnAff=fSnAff=f![$aA [R "SU5 SSKnUR)5nUR+U5 Us$f=f!,(df  g=f![a}n[5U5n S U ;a%S R7WU [8R:S5nOS R7WU 5n[R"U5 [[RU5eSnAff=f) aGLoads either a .joblib or .pkl file from GCS or from local. Loads one of DEFAULT_MODEL_FILE_NAME_JOBLIB or DEFAULT_MODEL_FILE_NAME_PICKLE files if they exist. This is used for both sklearn and xgboost. Arguments: model_path: The path to the directory that contains the model file. This path can be either a local path or a GCS path. Raises: PredictionError: If there is a problem while loading the file. Returns: A loaded scikit-learn or xgboost predictor object or None if neither DEFAULT_MODEL_FILE_NAME_JOBLIB nor DEFAULT_MODEL_FILE_NAME_PICKLE files are found. zgs://r)joblibNzCould not import joblib module.zLoading model %s using joblib.zRLoading model %s using joblib failed. Loading model using xgboost.Booster instead.zLoading model %s using pickle.rbzunsupported pickle protocolzCould not load the model: {}. {}. Please make sure the model was exported using python {}. Otherwise, please specify the correct 'python_version' parameter when deploying the model.z!Could not load the model: {}. {}.) startswithrLOCAL_MODEL_PATHrrrDEFAULT_MODEL_FILE_NAME_JOBLIBDEFAULT_MODEL_FILE_NAME_PICKLErsklearn.externalsr Exceptionrrrr8infoloadKeyErrorr Booster load_modelopenpickleloadsreadstrformatsys version_info) model_pathmodel_file_name_joblibmodel_file_name_picklemodel_file_namere error_msgr boosterf raw_error_msgs r%load_joblib_or_pickle_modelrs*$7## $45!J7KWW\\**HJWW\\**HJ ww~~,--.o Q -  5G{{?++ . / /.o ll3_E  &!||AFFH% ' & =Q Q - Q7)   I & D DiP P Q Q  (*9 ; //#?+  ' &   KFM$ 5 ABH&}c.>.>q.ACC 6<< =*i i />> JJ KsA.HE$'F$ A H$G29 H F!EH F"2FFFF!!H$AG/,H.G//H2 H<HH J  A8JJ cNS[U5R;a[$S[U5R;a[$SR [U5R[U5R 5n[ R"U5 [[RU5e)zDistinguish scikit-learn and xgboost using model object. Arguments: model_obj: A loaded model object Raises: PredictionError: If there is a problem detecting framework from object. Returns: Either scikit-learn framework or xgboost framework sklearnr z|Invalid model type detected: {}.{}. Please make sure the model file is an exported sklearn model, xgboost model or pipeline.) typer4SCIKIT_LEARN_FRAMEWORK_NAMEXGBOOST_FRAMEWORK_NAMErr3rcriticalrr8) model_objrs r% detect_sk_xgb_framework_from_objrs$y/,,, &&DO... !! %&,f O & & O $ $'&  Y />> JJr(cSnUHKn[RR[RRX55(dMFUS- nMM U$)aCount how many specified files exist in model_path. Args: model_path: The local path to the directory that contains the model file. specified_file_names: The file names to be checked Returns: An integer indicating how many specified_file_names are found in model_path. rr)rrrr)rspecified_file_names num_matches file_names r%_count_num_files_in_pathrsE+'i ww~~bggll:9::Qk( r(c[U[5n[U[5n[U[5nX-U-nUS:aASR U5n[ R "U5 [[RU5eUS:Xa[$US:Xa[$US:Xa[U5n[U5$[ R"S5 g)aSDetect framework from model_path by analyzing file extensions. Args: model_path: The local path to the directory that contains the model file. Raises: PredictionError: If framework can not be identified from model path. Returns: A string representing the identified framework or None (custom code is assumed in this situation). rz4Multiple model files are found in the model_path: {}zFModel files are not found in the model_path.Assumed to be custom code.N)r$TENSORFLOW_SPECIFIC_MODEL_FILE_NAMES!XGBOOST_SPECIFIC_MODEL_FILE_NAMESSCIKIT_LEARN_MODEL_FILE_NAMESrrrrr8rrrrwarning)rnum_tensorflow_modelsnum_xgboost_modelsnum_sklearn_modelsrrrs r%detect_frameworkr s368/35/ 0MO&:=OO+1_FMMI Y />> JJa $$Q !!Q+J7I +I 66 OO24 r(c[RRS5(dg[R"[RRS55nU(a[ U[ 5(dgURS5nU(a[ U[ 5(dg[R"SURU5U5 URU5$)aKGets the value of field_name in the version being created, if it exists. Args: field_name: Name of the key used for retrieving the corresponding value from version json object. Returns: The value of the given field in the version object or the user provided create version request if it exists. Otherwise None is returned. create_version_requestNversionz:Found value: %s, for field: %s from create_version_request) renvirongetjsonrrrrr) field_namerequestrs r%get_field_in_version_jsonr1s 0 1 1  JJrzz~~&>? @'  7D11  KK "'  7D11  ,,K{{:& 4 Z  r(c[U[R5(d#[SR [ U555e[ U;a([SR [ [ U555eUR[ 5$)zParses the predictions from the json response from prediction server. Args: response_json(Text): The JSON formatted response to parse. Returns: Predictions from the response json. Raises: ValueError if response_json is malformed. 4Invalid response received from prediction server: {}=Required field '{}' missing in prediction server response: {})rcollections_libMapping ValueErrorrreprPREDICTIONS_KEYpop response_jsons r%parse_predictionsrJs| M?#:#: ; ; >EE   ! ""M) GNN T-0 2 33   ? ++r(c[U[R5(d#[SR [ U555e[ U;a([SR [ [ U555eUR[ 5$)zParses the outputs from the json response from prediction server. Args: response_json(Text): The JSON formatted response to parse. Returns: Outputs from the response json. Raises: ValueError if response_json is malformed. rr)rrr r rr  OUTPUTS_KEYr rs r% parse_outputsras| M?#:#: ; ; >EE   ! "" % GNN m, . //   ; ''r(c[U[R5(d#[SR [ U555e[ U;a([SR [ [ U555eUR[ 5$)zParses instances from the json request sent to prediction server. Args: request_json(Text): The JSON formatted request to parse. Returns: Instances from the request json. Raises: ValueError if request_json is malformed. z-Invalid request sent to prediction server: {}zr-s0     .fOO/ .  ., "( .0"0!,($ #"!%3!   1( & 0 $ & * !,,.0   !Ni!NH  1%F1%h$D$6SSlP, 1*(:VMK`K8$%P!2,.(.)r(