B îq\.*ã@s–dZddlZddlZddlmZmZddlmZmZddl Z ddl m Z ddl mZddl mZdd lmZd d gZd gZdd d „ZGdd „d ƒZdS)z(General purpose timer related functions.éN)ÚIterableÚ OrderedDict)ÚpartialÚwraps)Úunits)Úlog)Úmodelingé)ÚAstropyUserWarningÚtimefuncÚRunTimePredictorTcs‡‡fdd„}|S)aUDecorator to time a function or method. Parameters ---------- num_tries : int, optional Number of calls to make. Timer will take the average run time. verbose : bool, optional Extra log information. Returns ------- tt : float Average run time in seconds. result Output(s) from the function. Examples -------- To add timer to time `numpy.log` for 100 times with verbose output:: import numpy as np from astropy.utils.timer import timefunc @timefunc(100) def timed_log(x): return np.log(x) To run the decorated function above: >>> t, y = timed_log(100) INFO: timed_log took 9.29832458496e-06 s on AVERAGE for 100 call(s). [...] >>> t 9.298324584960938e-06 >>> y 4.6051701859880918 cstˆƒ‡‡‡fdd„ƒ}|S)NcsZt ¡}xtˆƒD]}ˆ||Ž}qWt ¡}||ˆ}ˆrRt d ˆj|ˆ¡¡||fS)Nz*{0} took {1} s on AVERAGE for {2} call(s).)ÚtimeÚrangerÚinfoÚformatÚ__name__)ÚargsÚkwargsZtsÚiÚresultZteZtt)ÚfunctionÚ num_triesÚverbose©ú2lib/python3.7/site-packages/astropy/utils/timer.pyÚwrapperBs z1timefunc..real_decorator..wrapper)r)rr)rr)rrÚreal_decoratorAs z timefunc..real_decoratorr)rrrr)rrrr s* c@sdeZdZdZdd„Zedd„ƒZedddd d „ƒZd d „Z d d„Z ddd„Z dd„Z ddd„Z dS)r a©Class to predict run time. .. note:: Only predict for single varying numeric input parameter. Parameters ---------- func : function Function to time. args : tuple Fixed positional argument(s) for the function. kwargs : dict Fixed keyword argument(s) for the function. Examples -------- >>> from astropy.utils.timer import RunTimePredictor Set up a predictor for :math:`10^{x}`: >>> p = RunTimePredictor(pow, 10) Give it baseline data to use for prediction and get the function output values: >>> p.time_func(range(10, 1000, 200)) >>> for input, result in sorted(p.results.items()): ... print("pow(10, {0})\n{1}".format(input, result)) pow(10, 10) 10000000000 pow(10, 210) 10000000000... pow(10, 410) 10000000000... pow(10, 610) 10000000000... pow(10, 810) 10000000000... Fit a straight line assuming :math:`\text{arg}^{1}` relationship (coefficients are returned): >>> p.do_fit() # doctest: +SKIP array([1.16777420e-05, 1.00135803e-08]) Predict run time for :math:`10^{5000}`: >>> p.predict_time(5000) # doctest: +SKIP 6.174564361572262e-05 Plot the prediction: >>> p.plot(xlabeltext='Power of 10') # doctest: +SKIP .. image:: /_static/timer_prediction_pow10.png :width: 450px :alt: Example plot from `astropy.utils.timer.RunTimePredictor` When the changing argument is not the last, e.g., :math:`x^{2}`, something like this might work: >>> p = RunTimePredictor(lambda x: pow(x, 2)) >>> p.time_func([2, 3, 5]) >>> sorted(p.results.items()) [(2, 4), (3, 9), (5, 25)] cOsH|j|_t|f|ž|Ž|_tƒ|_g|_tƒ|_tƒ|_d|_ d|_ dS)N) rÚ _funcnamerÚ_pfuncrÚ _cache_goodÚ _cache_badÚ _cache_estÚ _cache_outÚ _fit_funcÚ_power)ÚselfÚfuncrrrrrÚ__init__–szRunTimePredictor.__init__cCs|jS)z¨Function outputs from `time_func`. A dictionary mapping input arguments (fixed arguments are not included) to their respective output values. )r")r%rrrÚresults szRunTimePredictor.resultsr F)rrcCs | |¡S)z1Run partial func once for single arg and time it.)r)r%ÚargrrrÚ _timed_pfuncªszRunTimePredictor._timed_pfuncc Cs‚||jkr~||jkr~y| |¡}Wn>tk r`}z t t|ƒt¡|j |¡Wdd}~XYnX|d|j|<|d|j |<dS)z(Cache timing results without repetition.Nrr ) rr r*Ú ExceptionÚwarningsÚwarnÚstrr Úappendr")r%r)rÚerrrÚ _cache_time¯szRunTimePredictor._cache_timecCs,t|tƒs|g}x|D]}| |¡qWdS)aETime the partial function for a list of single args and store run time in a cache. This forms a baseline for the prediction. This also stores function outputs in `results`. Parameters ---------- arglist : list of numbers List of input arguments to time. N)Ú isinstancerr1)r%Zarglistr)rrrÚ time_func»s  zRunTimePredictor.time_funcNécCsÌ||_tƒ|_t t|j ¡ƒ¡}|j|kr>t d  ||j¡ƒ‚|dkrTt j   d¡}n t|t jjƒstt j d  |¡¡‚|dkrˆt j ¡}n t|t jjƒs¨t j d  |¡¡‚||||t|j ¡ƒƒ|_|jjS)a5Fit a function to the lists of arguments and their respective run time in the cache. By default, this does a linear least-square fitting to a straight line on run time w.r.t. argument values raised to the given power, and returns the optimal intercept and slope. Parameters ---------- model : `astropy.modeling.Model` Model for the expected trend of run time (Y-axis) w.r.t. :math:`\text{arg}^{\text{power}}` (X-axis). If `None`, will use `~astropy.modeling.polynomial.Polynomial1D` with ``degree=1``. fitter : `astropy.modeling.fitting.Fitter` Fitter for the given model to extract optimal coefficient values. If `None`, will use `~astropy.modeling.fitting.LinearLSQFitter`. power : int, optional Power of values to fit. min_datapoints : int, optional Minimum number of data points required for fitting. They can be built up with `time_func`. Returns ------- a : array-like Fitted `~astropy.modeling.FittableModel` parameters. Raises ------ ValueError Insufficient data points for fitting. ModelsError Invalid model or fitter. zrequires {0} points but has {1}Nr z{0} is not a model.z{0} is not a fitter.)r$rr!ÚnpÚarrayÚlistrÚkeysÚsizeÚ ValueErrorrrZmodelsZ Polynomial1Dr2ZcoreZModelZfittingZ ModelsErrorZLinearLSQFitterZFitterÚvaluesr#Z parameters)r%ZmodelZfitterZpowerZmin_datapointsÚx_arrrrrÚdo_fitÐs&+     zRunTimePredictor.do_fitcCsF||jkr|j|}n,|jdkr(tdƒ‚| ||j¡}||j|<|S)aªPredict run time for given argument. If prediction is already cached, cached value is returned. Parameters ---------- arg : number Input argument to predict run time for. Returns ------- t_est : float Estimated run time for given argument. Raises ------ RuntimeError No fitted data for prediction. Nzno fitted data for prediction)r!r#Ú RuntimeErrorr$)r%r)Zt_estrrrÚ predict_times    zRunTimePredictor.predict_timeÚlinearrÚcsºddlm}tˆjƒ}t ‡fdd„|Dƒ¡}t|ƒdkrBtdƒ‚| ¡t j }xDt j t j t j t j t jfD](} | | ¡} d| krŒdkrjnqjPqjW|t j  | ¡}| ¡\} } | j||dd d ˆjdk rRtˆj ¡ƒ} t tˆj ¡ƒ¡t j  | ¡}| j| |d d d dt t|| ƒ¡}ˆ |ˆj¡t j  | ¡}| j||ddd |  |¡|  |¡|  |¡|  d |  ¡¡¡|  ˆj ¡| j!ddd| "¡|r¶| #|¡dS)aÒPlot prediction. .. note:: Uses `matplotlib `_. Parameters ---------- xscale, yscale : {'linear', 'log', 'symlog'} Scaling for `matplotlib.axes.Axes`. xlabeltext : str, optional Text for X-label. save_as : str, optional Save plot as given filename. Raises ------ RuntimeError Insufficient data for plotting. rNcsg|]}ˆj|‘qSr)r)Ú.0Úx)r%rrú Lsz)RunTimePredictor.plot..r zinsufficient data for plottingièzkx-ZActual)ÚlabelÚoÚrZ Predicted)ZmarkerÚcrEzb--ZFitz Run time ({})Zbest)ZlocZ numpoints)$Zmatplotlib.pyplotZpyplotÚsortedrr5r6Úlenr>ZmeanÚuÚsecondZminuteZ millisecondZ microsecondZ nanosecondZto_valueZsubplotsÚplotr#r7r!r8r;Zscatterr$Z set_xscaleZ set_yscaleZ set_xlabelZ set_ylabelrZ to_stringZ set_titlerZlegendZdrawZsavefig)r%ZxscaleZyscaleZ xlabeltextZsave_asZpltr<Zy_arrZqmeanZcur_uÚvalZfigZaxZx_estZy_estZx_fitZy_fitr)r%rrM1s@           zRunTimePredictor.plot)NNr r4)r@r@rrA)rÚ __module__Ú __qualname__Ú__doc__r'Úpropertyr(r r*r1r3r=r?rMrrrrr QsD   D)r T)rQr r,Ú collectionsrrÚ functoolsrrZnumpyr5ZastropyrrKrrÚ exceptionsr Ú__all__Z__doctest_skip__r r rrrrÚs     :