%g$ZddlmZmZmZddlZddlmZddlmZddl Z ddl Z ddl ZejdZdZddZdZdd Zdd Zd Zd Zd ZdZdZdZdZdZdZy))absolute_importdivisionprint_functionN)tqdm)deepcopyprophetc|dj|z }||djkr td|g}|dt|d|zk\r||z}|d|kD|d||zkzjs7||djkDr!||d|kjd}||z }|j ||dt|d|zk\r|dd}t |dk(r tdt jdjt ||d|dtt|S)aKGenerate cutoff dates Parameters ---------- df: pd.DataFrame with historical data. horizon: pd.Timedelta forecast horizon. initial: pd.Timedelta window of the initial forecast period. period: pd.Timedelta simulated forecasts are done with this period. Returns ------- list of pd.Timestamp dszLess data than horizon.NrzMLess data than horizon after initial window. Make horizon or initial shorter.z2Making {} forecasts with cutoffs between {} and {}) maxmin ValueErroranyappendlenloggerinfoformatlistreversed)dfhorizoninitialperiodcutoffresult closest_dates P/var/www/dash_apps/app1/venv/lib/python3.12/site-packages/prophet/diagnostics.pygenerate_cutoffsrsXX\\^g %F 4 233XF *BtH / /&T(V#4FW4D(DEJJL4 &!"T(f"4599;DA %/ f *BtH / /CR[F 6{a /   KKDKK F VBZ  !!c  j tdjjjdt j ddgj rjddg|7t|tr|g}j|Dcgc] }|vs| c}d } jjD]} t| | d } t j t| d z} |Q|d znt j |}|td z| nt j |}t||}nrt|djkr tddjz } t|| kDr td|ddjz }|| kr0dj!| } | dz } | dz } t"j%| |r hd}|dk(rt&j(j+}n|dk(rt&j(j-}nd|dk(r% ddlm}|}|j5g\n:t7|dr|}n+dj!dj9|} t| fd|D}t;|}t"j=d||j>t@g|}|dk(r;|jC|}n)|s tE|n|Dcgc]}tA|}}t jF|djdScc}w#t2$r}t3d|d}~wwxYwcc}w) aO Cross-Validation for time series. Computes forecasts from historical cutoff points, which user can input. If not provided, begins from (end - horizon) and works backwards, making cutoffs with a spacing of period until initial is reached. When period is equal to the time interval of the data, this is the technique described in https://robjhyndman.com/hyndsight/tscv/ . Parameters ---------- model: Prophet class object. Fitted Prophet model. horizon: string with pd.Timedelta compatible style, e.g., '5 days', '3 hours', '10 seconds'. period: string with pd.Timedelta compatible style. Simulated forecast will be done at every this period. If not provided, 0.5 * horizon is used. initial: string with pd.Timedelta compatible style. The first training period will include at least this much data. If not provided, 3 * horizon is used. cutoffs: list of pd.Timestamp specifying cutoffs to be used during cross validation. If not provided, they are generated as described above. parallel : {None, 'processes', 'threads', 'dask', object} How to parallelize the forecast computation. By default no parallelism is used. * None : No parallelism. * 'processes' : Parallelize with concurrent.futures.ProcessPoolExectuor. * 'threads' : Parallelize with concurrent.futures.ThreadPoolExecutor. Note that some operations currently hold Python's Global Interpreter Lock, so parallelizing with threads may be slower than training sequentially. * 'dask': Parallelize with Dask. This requires that a dask.distributed Client be created. * object : Any instance with a `.map` method. This method will be called with :func:`single_cutoff_forecast` and a sequence of iterables where each element is the tuple of arguments to pass to :func:`single_cutoff_forecast` .. code-block:: class MyBackend: def map(self, func, *iterables): results = [ func(*args) for args in zip(*iterables) ] return results disable_tqdm: if True it disables the progress bar that would otherwise show up when parallel=None extra_output_columns: A String or List of Strings e.g. 'trend' or ['trend']. Additional columns to 'yhat' and 'ds' to be returned in output. Returns ------- A pd.DataFrame with the forecast, actual value and cutoff. Nz^Model has not been fit. Fitting the model provides contextual parameters for cross validation.Tdropr yhat yhat_lower yhat_uppergrz daysg?zEMinimum cutoff value is not strictly greater than min date in historyzdMaximum cutoff value is greater than end date minus horizon, no value for cross-validation remainingrz"Seasonality has period of {} days z%which is larger than initial window. zConsider increasing initial.>daskthreads processesr)r*r() get_clientz6parallel='dask' requires the optional dependency dask.mapzB'parallel' should be one of {} for an instance with a 'map' methodz, c3,K|] }|f ywN).0rrrmodelpredict_columnss r z#cross_validation..s$,%/B,szApplying in parallel with %saxis)$history Exceptioncopy reset_indexpd Timedeltauncertainty_samplesextend isinstancestr seasonalitiesvaluesr rr rrrwarning concurrentfuturesThreadPoolExecutorProcessPoolExecutordask.distributedr+ ImportErrorscatterhasattrjoinziprr,single_cutoff_forecastgatherrconcat)r1rrrparallelcutoffs disable_tqdmextra_output_columnsc period_maxsseasonality_dtend_date_minus_horizonmsgvalidpoolr+e iterablespredictsrrr2s`` @@rcross_validationr_=syv }}xyy     ) )t ) 4Bll7#GVnO    l;<' *C 0$8#9 +?\a1OC[\]J  ' ' )2Qx[1 2\\#j/G"; )de e!#D'!9 w<0 0DE E!*r$x||~- 6==jIC : :C 1 1C NN3 0 y %%88:D  $%%99;D   =7- M99M>c t||}||d|k}|jddkr td|j|fi|j|d|kD|d||zkz}dg}|j dk(r.|j d|jr|j d|j|jj|j|jjD cgc] } | d| dc} |j|||} tj| |||d gj!d tj"d |gt%| zigd Scc} w)aForecast for single cutoff. Used in cross validation function when evaluating for multiple cutoffs either sequentially or in parallel . Parameters ---------- df: pd.DataFrame. DataFrame with history to be used for single cutoff forecast. model: Prophet model object. cutoff: pd.Timestamp cutoff date. Simulated Forecast will start from this date. horizon: pd.Timedelta forecast horizon. predict_columns: List of strings e.g. ['ds', 'yhat']. Columns with date and forecast to be returned in output. Returns ------- A pd.DataFrame with the forecast, actual value and cutoff. r rz@Less than two datapoints before cutoff. Increase initial window.logisticcapfloorcondition_nameyTr"rr4) prophet_copyshaper7fit fit_kwargsgrowthrlogistic_floorr=extra_regressorskeysr@rApredictr:rOr9 DataFramer) rr1rrr2m history_cindex_predictedcolumnspropsr$s rrMrMs. UF#A2d8v%&IqA '   AEE)(u''($x&(RX'9I-IJOfGxx:u   NN7 # NN1%%**,- NN__++-0  ! " . 01 99R(1 2D 99 _ ?SE"..D.9 h3t9 456   0s6E;c|j td|jrB|j}|6t |jd|jd|k}|||k}nd}|j |j |j|j|ddd|j|j|j|j|j|j|j|j |j"|j$|j$j'nd}t)|j*|_t)|j,|_t)|j.|_|S)aZCopy Prophet object Parameters ---------- m: Prophet model. cutoff: pd.Timestamp or None, default None. cuttoff Timestamp for changepoints member variable. changepoints are only retained if 'changepoints <= cutoff' Returns ------- Prophet class object with the same parameter with model variable Nz,This is for copying a fitted Prophet object.r F)rln_changepointschangepoint_range changepointsyearly_seasonalityweekly_seasonalitydaily_seasonalityholidays holidays_modeseasonality_modeseasonality_prior_scalechangepoint_prior_scaleholidays_prior_scale mcmc_samplesinterval_widthr< stan_backend)r6r7specified_changepointsrzr __class__rlrxryr~rrrrrrrr<rget_typerrnr@country_holidays)rrrrzlast_history_datem2s rrhrhsK yyFGG~~   #AIIdOAIIdOv4M$N O ' 7H(HIL  xx''--!  oo++ ! 9 9 ! 9 933^^''11)*)CANN # # %'  B,#1#5#56B0B"1#5#56B Ir cgd}||}d|vsd|vrd|vr|jdtt|t|k7r tdt|j t|stdj ||j }|rg|djjd jt|d jjd jtz |d <n|d|d z |d <|jd d d|vrJ|djjdkr&tjd|jdt|dk(ryt||j dz}|dk\r%t#|d}t||j d}i}|D]}t%|||||<||d} t'dt|D]J} ||| } t)j*| d j,| d j,sJ| || | || <L| S)aCompute performance metrics from cross-validation results. Computes a suite of performance metrics on the output of cross-validation. By default the following metrics are included: 'mse': mean squared error 'rmse': root mean squared error 'mae': mean absolute error 'mape': mean absolute percent error 'mdape': median absolute percent error 'smape': symmetric mean absolute percentage error 'coverage': coverage of the upper and lower intervals A subset of these can be specified by passing a list of names as the `metrics` argument. Metrics are calculated over a rolling window of cross validation predictions, after sorting by horizon. Averaging is first done within each value of horizon, and then across horizons as needed to reach the window size. The size of that window (number of simulated forecast points) is determined by the rolling_window argument, which specifies a proportion of simulated forecast points to include in each window. rolling_window=0 will compute it separately for each horizon. The default of rolling_window=0.1 will use 10% of the rows in df in each window. rolling_window=1 will compute the metric across all simulated forecast points. The results are set to the right edge of the window. If rolling_window < 0, then metrics are computed at each datapoint with no averaging (i.e., 'mse' will actually be squared error with no mean). The output is a dataframe containing column 'horizon' along with columns for each of the metrics computed. Parameters ---------- df: The dataframe returned by cross_validation. metrics: A list of performance metrics to compute. If not provided, will use ['mse', 'rmse', 'mae', 'mape', 'mdape', 'smape', 'coverage']. rolling_window: Proportion of data to use in each rolling window for computing the metrics. Should be in [0, 1] to average. monthly: monthly=True will compute horizons as numbers of calendar months from the cutoff date, starting from 0 for the cutoff month. Returns ------- Dataframe with a column for each metric, and column 'horizon' )msermsemaemapemdapesmapecoverageNr%r&rz-Input metrics must be a list of unique valuesz Valid values for metrics are: {}r MrrT)inplacerrfg:0yE>z"Skipping MAPE because y close to 0rrg)removersetrissubsetrr8dt to_periodastypeint sort_valuesabsr rrrir evalrangenp array_equalrA) rmetricsrolling_windowmonthly valid_metricsdf_mwdfsmetricresires_ms rperformance_metricsrFs<^QMB,b"8zW?Tz" 3w<CL(HII w< ]!3 4 . 5 5m D   779Dt*--11#6==cBT(^EVEVE`E`adEeElElmpEqqYt*tH~5YY- T#Y]]_002T9 89v 7|q NTZZ] *+AAv 1I 4::a= ! C,"d6l4+F , gaj/C 1c'l #,GAJ~~c)n33U95E5L5LMMM +GAJ, Jr ctj||d}|jdjddgj j d}|ddj }|ddj }|jj }t|dz } d} d} tjt|} tt|dz ddD]Q} | || z } | || z } | |k\s| |z }||| z|| z }| |z |z | | <| || z} | || z} | dz} | |k\r9S|| dzd }| | dzd } tjd ||| iS) aCompute a rolling mean of x, after first aggregating by h. Right-aligned. Computes a single mean for each unique value of h. Each mean is over at least w samples. Parameters ---------- x: Array. h: Array of horizon for each value in x. w: Integer window size (number of elements). name: Name for metric in result dataframe Returns ------- Dataframe with columns horizon and name, the rolling mean of x. xhrsumcountrrgrr Nr) r:rqgroupbyaggr9rrArrremptyr)rrrnamerdf2xsnshs trailing_ix_sumn_sumres_xrexcess_nexcess_xres_hs rrolling_mean_by_hrs$ AA& 'B 3UG,-99;GGL S%  B S'  ! !B BSAJ E E HHSX E3s8a<R (  A AqjqyH"Q%'"Q%/H!&!1A 5E*  R ^ #E R ^ #E !OJqj   Q !E :># $E <<E47 88r c*tj||d}|jd}|jj j d}|d}g}g} t |dz } | dk\r|| } |j| jj} tj|| k(jdz } t | |kr2| dk\r-| j|| | dz} t | |kr| dk\r-t | |krnC|j|| | jtj| | dz} | dk\r|j| jtjd||| iS)aCompute a rolling median of x, after first aggregating by h. Right-aligned. Computes a single median for each unique value of h. Each median is over at least w samples. For each h where there are fewer than w samples, we take samples from the previous h, moving backwards. (In other words, we ~ assume that the x's are shuffled within each h.) Parameters ---------- x: Array. h: Array of horizon for each value in x. w: Integer window size (number of elements). name: Name for metric in result dataframe Returns ------- Dataframe with columns horizon and name, the rolling median of x. rrrgrr)r:rqrsizer9rr get_grouprtolistrarrayargmaxrmedianreverse)rrrrrgroupedrrrrrh_irnext_idx_to_adds rrolling_median_by_hrsi* AA& 'BjjoG ,,. $ $ & 2 23 7C SB E E B! A q&e   s # % % , , .((18,33592w{A!5 IIa( ) q O 2w{A!5 r7Q;  RU RYYr]# Q! q&" MMO MMO <<E47 88r c|d|dz dz}|dkrtj|d|dSt|j|dj|dS) zMean squared error Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and mse. rfr$rarr)rrrrrrrr:rqrrA)rrses rrr s^ S'BvJ 1 $B1u||9 bABB  ))r)}++qu r ct||}tj|d|d<|jddidd|S)zRoot mean squared error Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and rmse. rrruT)r5r)rrsqrtrename)rrrs rrrs@ b!*CU$CJJJvYJ= Jr ctj|d|dz }|dkrtj|d|dSt |j |dj |dS)zMean absolute error Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and mae. rfr$rr)rrrrrrr:rqrrA)rraes rrr/sb 3"V*$ %B1u||9 bABB  ))r)}++qu r ctj|d|dz |dz }|dkrtj|d|dSt |j |dj |dS)zMean absolute percent error Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and mape. rfr$rr)rrrrrrrapes rrrCsk &&"S'BvJ&"S'1 2C1u||9 sCDD  **9 ,, r ctj|d|dz |dz }|dkrtj|d|dSt |j |d|dS)zMedian absolute percent error Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and mdape. rfr$rr)rrrr)rrr:rqrrArs rrrWse &&"S'BvJ&"S'1 2C1u||9 DEE  **9  r cVtj|d|dz tj|dtj|dzdz z }|jd}|dkrtj|d|dSt |j |dj |dS) a Symmetric mean absolute percentage error based on Chen and Yang (2004) formula Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and smape. rfr$rarr)rrrr)rrfillnar:rqrrA)rrsapes rrrks 66"S'BvJ& 'BFF2c7ObffRZ>P,PTU+U VD ;;q>D1u||9 EFF  ++I-- r c|d|dk\|d|dkz}|dkrtj|d|dSt|j|dj|dS) zCoverage Parameters ---------- df: Cross-validation results dataframe. w: Aggregation window size. Returns ------- Dataframe with columns horizon and coverage. rfr%r&rr)rrrrr)rr is_covereds rrrsoS'R --"S'R =M2MNJ1u||9 :NOO    r)}33qz r )NNNNFNr.)Ng?F) __future__rrrlogging tqdm.autorr8rconcurrent.futuresrCnumpyrpandasr: getLoggerrrr_rMrhrrrrrrrrrrr/r rrsA@   9 %&"RW>t4n5pRj19j19r($(((,r