M/PhddlZddlmZddlmZddlmZd dZddZ ddZ dd Z e e e d Z ed ee d ZdS)N)scoreatpercentile) Substitution)kernelscd}t|dt|dz |z }tj|dd}|dkrtj||S|S)z Returns the smaller of std(X, ddof=1) or normalized IQR(X) over axis 0. References ---------- Silverman (1986) p.47 g/$?Krr)axisddof)rnpstdminimum)x percentile normalizeIQRstd_devs d/var/www/html/test/jupyter/venv/lib/python3.11/site-packages/statsmodels/nonparametric/bandwidths.py _select_sigmarscI Q # #&72&>&> >) KCfQQQ'''G Qwwz'3'''cTt|}t|}d|z|dzzS)a* Scott's Rule of Thumb Parameters ---------- x : array_like Array for which to get the bandwidth kernel : CustomKernel object Unused Returns ------- bw : float The estimate of the bandwidth Notes ----- Returns 1.059 * A * n ** (-1/5.) where :: A = min(std(x, ddof=1), IQR/1.349) IQR = np.subtract.reduce(np.percentile(x, [75,25])) References ---------- Scott, D.W. (1992) Multivariate Density Estimation: Theory, Practice, and Visualization. gl?皙ɿrlenrkernelAns rbw_scottrs/: aA AA 19qT{ ""rcTt|}t|}d|z|dzzS)a Silverman's Rule of Thumb Parameters ---------- x : array_like Array for which to get the bandwidth kernel : CustomKernel object Unused Returns ------- bw : float The estimate of the bandwidth Notes ----- Returns .9 * A * n ** (-1/5.) where :: A = min(std(x, ddof=1), IQR/1.349) IQR = np.subtract.reduce(np.percentile(x, [75,25])) References ---------- Silverman, B.W. (1986) `Density Estimation.` g?rrrs r bw_silvermanr!<s/8 aA AA 6A$K rc|tj}|j}t|}t |}||z|dzzS)a{ Plug-in bandwidth with kernel specific constant based on normal reference. This bandwidth minimizes the mean integrated square error if the true distribution is the normal. This choice is an appropriate bandwidth for single peaked distributions that are similar to the normal distribution. Parameters ---------- x : array_like Array for which to get the bandwidth kernel : CustomKernel object Used to calculate the constant for the plug-in bandwidth. The default is a Gaussian kernel. Returns ------- bw : float The estimate of the bandwidth Notes ----- Returns C * A * n ** (-1/5.) where :: A = min(std(x, ddof=1), IQR/1.349) IQR = np.subtract.reduce(np.percentile(x, [75,25])) C = constant from Hansen (2009) When using a Gaussian kernel this is equivalent to the 'scott' bandwidth up to two decimal places. This is the accuracy to which the 'scott' constant is specified. References ---------- Silverman, B.W. (1986) `Density Estimation.` Hansen, B.E. (2009) `Lecture Notes on Nonparametrics.` Nr)rGaussiannormal_reference_constantrr)rrCrrs rbw_normal_referencer&]sKN~!##(AaA AA q51; r)scott silvermannormal_referencez, c|}|tvrtd|zt|||}tj|dkrd}t ||S)a Selects bandwidth for a selection rule bw this is a wrapper around existing bandwidth selection rules Parameters ---------- x : array_like Array for which to get the bandwidth bw : str name of bandwidth selection rule, currently supported are: %s kernel : not used yet Returns ------- bw : float The estimate of the bandwidth zBandwidth %s not understoodrzSelected KDE bandwidth is 0. Cannot estimate density. Either provide the bandwidth during initialization or use an alternative method.)lowerbandwidth_funcs ValueErrorr any RuntimeError)rbwr bandwidtherrs rselect_bandwidthr3sw* B   6;<<<#Av..I vi1n'3r)r)N)numpyr scipy.statsrstatsmodels.compat.pandasr!statsmodels.sandbox.nonparametricrrrr!r&r,joinsortedkeysr3rrr<s))))))222222555555&####B    B,,,,j+dii335566778898r