How to Resample Time Series Data in Python?

In time series, data consistency is of prime importance, resampling ensures that the data is distributed with a consistent frequency. Resampling can also provide a different perception of looking at the data, in other words, it can add additional insights about the data based on the resampling frequency.

resample() function: It is a  primarily used for time series data.

Syntax:

# import the python pandas library
import pandas as pd

# syntax for the resample function.
pd.series.resample(rule, axis=0, closed='left',
 convention='start', kind=None, offset=None,
 origin='start_day')

Resampling primarily involves changing the time-frequency of the original observations. The two popular methods of resampling in time series are as follows

• Upsampling
• Downsampling

Upsampling

Upsampling involves increasing the time-frequency of the data, it is a data disaggregation procedure where we break down the time frequency from a higher level to a lower level. For example Breaking down the time-frequency from months to days, or days to hours or hours to seconds. Upsampling usually blows up the size of the data, depending on the sampling frequency. If D is the size of original data and D’ is the size of Upsampled data, then D’ > D

Now, let’s look at an example using Python to perform resampling in time-series data.

Click here to download the practice dataset Detergent sales data.csv used for the implementation.

Example:

Output:

The detergent sales data shows sales value for the first 6 months. Assume the task here is to predict the value of the daily sales. Given monthly data, we are asked to predict the daily sales data, which signifies the use of Upsampling. 

Output:

The output shows a few samples of the dataset which is upsampled from months to days, based on the mean value of the month. You can also try using sum(), median() that best suits the problem.

The dataset has been upsampled with nan values for the remaining days except for those days which were originally available in our dataset. (total sales data for each month).

Now, we can fill these nan values using a technique called Interpolation. Pandas provide a function called DataFrame.interpolate() for this purpose. Interpolation is a method that involves filling the nan values using one of the techniques like nearest’, ‘zero’, ‘slinear’, ‘quadratic’, ‘cubic’, ‘spline’, ‘barycentric’, ‘polynomial’. We will choose “linear” interpolation. This draws a straight line between available data, in this case on the last of the month, and fills in values at the chosen frequency from this line. 

Output:

Upsampling with a polynomial interpolation

Another common interpolation method is to use a polynomial or a spline to connect the values. This creates more curves and can look realistic on many datasets. Using a spline interpolation requires you to specify the order (number of terms in the polynomial).

Output:

Thus, we can use resample() and interpolate() function to upsample the data. Try this out using different configurations of these functions.

Downsampling:

Downsampling involves decreasing the time-frequency of the data, it is a data aggregation procedure where we aggregate the time frequency from a lower level to a higher level. For example summarizing the time-frequency from days to months, or hours to days or seconds to hours. Downsampling usually shrinks the size of the data, depending on the sampling frequency. If D is the size of original data and D’ is the size of Upsampled data, then D’ < D.

For example, car sales data shows sales value for the first 6 months daywise. Assume the task here is to predict the value of the quarterly sales. Given daily data, we are asked to predict the quarterly sales data, which signifies the use of downsampling. 

Click here to download the practice dataset car-sales.csv used in this implementation.

Example:

Output:

We can use quarterly resampling frequency ‘Q’ to aggregate the data quarter-wise.

Output:

Now, this downsampled data can be used for predicting quarterly sales.