Part 4: Time Series Analysis in Pandas

In [ ]:

    

"""
----------------------------------------------------------------------
Filename : 04_time_series_in_pandas.py
Date     : 12th Dec, 2013
Author   : Jaidev Deshpande
Purpose  : Introduction to time series analysis and plotting in Pandas
Libraries: Pandas, NumPy, Matplotlib
----------------------------------------------------------------------
"""


    

In [ ]:

    

# standard library imports
from datetime import date

# system library imports
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt


    

In [ ]:

    

def create_timestamp_sample():
    """
    Create a Pandas `TimeSeries` object with random data, 
    timestamped from 1st January 2000 to 31st December 2013
    """
    times = pd.date_range(start='1/1/2000', end='31/12/2013')
    data = np.random.randn(len(times))
    return pd.Series(data, index=times)


    

In [ ]:

    

ts = create_timestamp_sample()


    

In [ ]:

    

print(ts['1/1/2000'])


    

In [ ]:

    

# Create a datetime object
dt = date(2000,1,1)
print(ts[dt])


    

In [ ]:

    

# Slicing a timeseries by dates:
print(ts['1/1/2000':'31/1/2000'])


    

In [ ]:

    

# Changing the frequency of a TimeSeries
converted_s = ts.asfreq('M',method='bfill')
fig, axes = plt.subplots(nrows=2,ncols=1)
ts.plot(ax=axes[0],figsize=(20,10))
axes[0].set_title('Daily, for 14 years')
converted_s.plot(ax=axes[1])
axes[1].set_title('Monthly, for 14 years')
plt.show()


    

In [ ]:

    

def series_interpolation(n_sample):
    # Create a Series with some empty data
    x = np.linspace(-4*np.pi, 4*np.pi, 1000)
    y = np.sin(x)
    samples = np.arange(0,1000,1000.0/n_sample).astype(int)
    y_sampled = np.zeros((1000,),float)
    y_sampled[samples] = y[samples]
    y_sampled[y_sampled==0] = np.nan
    Y = pd.Series(y_sampled)
    return Y


    

In [ ]:

    

# Series interpolation demo
sample_sizes = [10,25,50,75,100]
sampled_data = [series_interpolation(n) for n in sample_sizes]
fig, axes = plt.subplots(nrows=len(sample_sizes),ncols=1, figsize=(20,10))
for i in range(len(sample_sizes)):
   data = sampled_data[i]
   interpolated = data.interpolate()
   data[data.notnull()].plot(style='ro',ax=axes[i])
   interpolated.plot(style='b-', ax=axes[i])
   axes[i].set_title("Sample Size = "+str(sample_sizes[i]))
plt.show()


    

In [ ]:

    

# Upsampling and downsampling
print(ts.index.freq)
upsampled = ts.resample('12H',fill_method='bfill')
downsampled = ts.resample('M')
print(upsampled.shape)
print(downsampled.shape)
fig, axes = plt.subplots(3,1, figsize=(20,10))
ts.plot(ax=axes[0],title='Original Data')
upsampled.plot(ax=axes[1],title='Upsampled Data')
downsampled.plot(ax=axes[2], title='Downsampled Data')
plt.show()


    

In [ ]:

    

def create_interpolation_exercise():
    random.seed(0)
    x = zeros((1000,),dtype=float)
    inds = random.randint(low=0, high=1000, size=(100,))
    values = random.randn(100)
    x[inds] = values
    x[x==0] = nan
    return pd.Series(x)