KDE peak properties (kdepeakprops)

---

Brief description

A module to get some basic properties of a KDE peak: peak location, FWHM and
its upper and lower bounds. The KDE is estimated with sklearn.neighbors.KernelDensity.

kdepeakprops.kde_props takes a 1D array of values and returns an
array with the KDE peak properties (value, FWHM, upp, low) and another (2D array) with the KDE samples.

---

Install

cd /path/to/install/dir
git clone https://github.com/iskren-y-g/kdepeakprops.git

Add path or

import sys
sys.path.insert(0,'/path/to/install/dir')

---

Example:

Some imports first:

import numpy as np
from matplotlib import pyplot as plt

import kdepeakprops as kpp

Generate some test data and plot it

np.random.seed(42)
x = np.append(np.random.normal(1900, 100, size=10000),
              np.random.normal(2120, 100, size=5000))

fig,ax = plt.subplots()
_ = ax.hist(x, bins=60)

---

Calculate the FWHM of the KDE peak

Use the auto plotting feature

kde_peak_prop,kde_xy = kpp.kde_props(x, nbins=60, show=True)

KDE estimation with Gaussian kernel width = 15.919
KDE peak = 1916.009 +235.197, -130.984

---

Plot the result manually

kde_peak_prop,kde_xy = kpp.kde_props(x, nbins=60, show=False)
peak, fwhm, hwhm_upp, hwhm_low = kde_peak_prop

label = r'X = {:.3f} $^{{+{:.3f}}}_{{-{:.3f}}}$, FWHM={:.3f}'.format(
    peak,hwhm_upp-peak,peak-hwhm_low, fwhm)

fig,ax = plt.subplots()
_ = ax.hist(x, bins=60, color='grey', alpha =.5)
ax.axvline(peak, color='C0')
ax.axvline(hwhm_low, color='C0')
ax.axvline(hwhm_upp, color='C0')
ax.hlines(kde_xy[1][kde_xy[1].argmax()]/2, 
          hwhm_low, hwhm_upp, color='C1')
ax.plot(kde_xy[0], kde_xy[1], '-', color='C0')
ax.text(ax.get_xlim()[0],
        ax.get_ylim()[1]*1.1, '{:s}'.format(label), color='C0')
ax.set_xlabel('X')
ax.set_ylabel('N')

KDE estimation with Gaussian kernel width = 15.919
KDE peak = 1916.009 +235.197, -130.984





Text(0, 0.5, 'N')

---

An example with another test data

Load test data as numpy array

import pickle
filename = 'test/data/example_sample.pickle'
with open(filename, 'rb') as f:
    x = pickle.load(f)

Print basic stats of the test data

_,_ = kpp.kde_props(x, show=True)

Using 'doane' metod to diplay 21 bins.
KDE estimation with Gaussian kernel width = 0.069
KDE peak = -1.615 +0.313, -0.156

Different example data

filename = 'test/data/example_data2.pickle'
with open(filename, 'rb') as f:
    x = pickle.load(f)

print('len = {:}, min = {:.2f}, max = {:.2f}\n 16%,50%,84% = {:}\n std_mean = {:}\
'.format(len(x), x.min(), x.max(), np.percentile(x,[15.9,50,84.1]), 
         x.std()/np.sqrt(len(x))
                                                                       )
     )

len = 10000, min = 128.75, max = 253418.62
 16%,50%,84% = [  990.73092484  3579.07433514 17294.38213141]
 std_mean = 167.9843211320867

Plot it

_ = plt.hist(x[x<6000],bins=100)

Analyse it with fwhm_from_kde

_,_ = kpp.kde_props(x[x<6000], nbins=100, show=True)

KDE estimation with Gaussian kernel width = 58.707
KDE peak = 816.313 +611.168, -229.188