Clever Geek Handbook

Is there a bokeh version for pandas autocorrelation method?

Let's say we have a time series object called "series" . I know its a very easy to use autocorrelation_plot () method for plotting the Lag and Autocorrelation parameters of a series object .

Here is the code:

from matplotlib import pyplot
from pandas.tools.plotting import autocorrelation_plot
autocorrelation_plot(series)
pyplot.show()

Here is a pandas graph:

enter image description here

Is there a way to get the same plot using bokeh server?

+3


source to share


2 answers


Yes there is. I wrote codes that give you the same output from the pandas autocorrelation_plot () method .

Here is the code:

from bokeh.layouts import column
from bokeh.plotting import figure, curdoc
import timeseries_model_creator  # to get data
import numpy as np

TimeSeriesModelCreator = timeseries_model_creator.TimeSeriesModelCreator()
series = TimeSeriesModelCreator.read_csv()  # time series object

def get_autocorrelation_plot_params(series):
    n = len(series)
    data = np.asarray(series)

    mean = np.mean(data)
    c0 = np.sum((data - mean) ** 2) / float(n)

    def r(h):
        return ((data[:n - h] - mean) *
                (data[h:] - mean)).sum() / float(n) / c0
    x = np.arange(n) + 1
    y = map(r, x)
    print "x : ", x, " y : ", y
    z95 = 1.959963984540054
    z99 = 2.5758293035489004
    return n, x, y, z95, z99

n, x, y, z95, z99 = get_autocorrelation_plot_params(series)

auto_correlation_plot2 = figure(title='Time Series Auto-Correlation', plot_width=1000,
                                plot_height=500, x_axis_label="Lag", y_axis_label="Autocorrelation")

auto_correlation_plot2.line(x, y=z99 / np.sqrt(n), line_dash='dashed', line_color='grey')
auto_correlation_plot2.line(x, y=z95 / np.sqrt(n), line_color='grey')
auto_correlation_plot2.line(x, y=0.0, line_color='black')
auto_correlation_plot2.line(x, y=-z95 / np.sqrt(n), line_color='grey')
auto_correlation_plot2.line(x, y=-z99 / np.sqrt(n), line_dash='dashed', line_color='grey')

auto_correlation_plot2.line(x, y, line_width=2)
auto_correlation_plot2.circle(x, y, fill_color="white", size=8)  # optional

curdoc().add_root(column(auto_correlation_plot2))


Here is the bokeh graph:

enter image description here

+6


source


I am basically copying Ayberk's answer, but turning it into a function

def acf(series):
    n = len(series)
    data = np.asarray(series)
    mean = np.mean(data)
    c0 = np.sum((data - mean) ** 2) / float(n)
    def r(h):
        acf_lag = ((data[:n - h] - mean) * (data[h:] - mean)).sum() / float(n) / c0
        return round(acf_lag, 3)
    x = np.arange(n) # Avoiding lag 0 calculation
    acf_coeffs = pd.Series(map(r, x)).round(decimals = 3)
    acf_coeffs = acf_coeffs + 0
    return acf_coeffs

def significance(series):
    n = len(series)
    z95 = 1.959963984540054 / np.sqrt(n)
    z99 = 2.5758293035489004 / np.sqrt(n)
    return(z95,z99)

def bok_autocor(series):
    x = pd.Series(range(1, len(series)+1), dtype = float)
    z95, z99 = significance(series)
    y = acf(series)
    p = figure(title='Time Series Auto-Correlation', plot_width=1000,
               plot_height=500, x_axis_label="Lag", y_axis_label="Autocorrelation")
    p.line(x, z99, line_dash='dashed', line_color='grey')
    p.line(x, z95, line_color = 'grey')
    p.line(x, y=0.0, line_color='black')
    p.line(x, z99*-1, line_dash='dashed', line_color='grey')
    p.line(x, z95*-1, line_color = 'grey')
    p.line(x, y, line_width=2)
    return p


show(series.pipe(bok_autocor))


I found acf here which says it is from the autocorrelation_plot pandas function and looks as good as Ayberk used.

+2


source






More articles:


All Articles