Design Choices

Here are outlined the design choices which have been taken during the development of the library.

TimeSeries Operations

Let’s say we have two timeseries objects ts1 and ts2 and need to obtain a new timeseries which contains as values the sum of the two:

ts = ts1+ts2

Mismatching dates

In situations where an operation is performed on two timeseries it can not assumed that the same dates are present in both series.

To allow for this there need to be suitable timeseries operators, which allow the user to specify the necessary settings. Where standard mathematical binary operators are used such as:

• • / *

suitable defaults will be assumed.

An example of a timeseries operators include:

>>>import operators
>>>ts = operators.add(ts1, ts2, all = True, fill = None)
>>>#all indicates all dates are included
...#fill indicates the value to use for missing value. It can be a primitive or a class
>>>

Proposed functionality

A timeseries should be able to get/set the value(s) for a given date. For example:

>>> from dynts import timeseries
>>> ts = timeseries('test')
>>> dt = datetime.date(2010,01,01)
>>> px = 101.1
>>> ts[dt] = px
>>> ts[dt]
>>> 101.1

It should be able to perform standard mathematical functions on them as if they were vectors. For example:

>>> ts2 = timeseries('test2')
>>> ts2[dt] = 10.0
>>> ts3 = ts1 + ts2
>>> ts3
>>> 2010-01-01 : 111.1

In the case where a date is missing from one of the series a suitable default/error value should be supplied:

>>> dt2 = datetime.date(2010, 01, 02)
>>> ts2[dt2] = 100.2
>>> ts3 = ts2 + ts1
>>> ts3
>>> 2010-01-01 : 111.1
... 2010-01-02 : None

Potentially , each back-end will have its own missing value, therefore a suitable conversion to a unique python missing value needs to be in place.

In situations where access to the values is required by index (rather than by date) there is an appropriate helper class:

>>> ts_as_matrix = as_matrix(ts3)
>>> ts_as_matrix[0]
>>> 111.1

This should be easy to do.

It may be useful to move functionality of the ts into a seperate ‘analytics’ object to avoid creating undue dependencies. Each function in the analytics object would return either a scalar value or another timeseries object with the same backend:

>>> scalar_sd = ts.analytics.stdeviation(window = None)
>>> ts_ma40 = ts.analytics.ma(window = 40)