Time series analysis

A time series is a set of data points ordered by time. For example, in such a data set, a sale in January 2020 comes before a sale in February 2020. You can use ThoughtSpot’s time series analysis feature to search for answers about series data.

Automatic time series suggestions

You might use this feature to compare a time period across other time periods. For example, you might look at sales for each month across several years. You may also want to calculate an aspect such as growth over the same time period across other periods. You can also do relative analysis, such as sales for the last 3 months of each year across several years.

You can use one or more of the following period keywords to create this type of analysis:

Period keywords

day
Example
orders by day
day of month
Example
sales day of month by month
day of quarter
Example
deals day of quarter by year
day of week
Example
customers by week day of week
day of year
Example
ad clicks day of year by year
hour
Example
visitors by hour weekly
month of quarter
Example
purchases month of quarter by year
quarter
Example
purchases by quarter
quarter of year
Example
opportunities quarter of year last 4 years yearly
week of month
Example
product shipments week of month yearly
week of quarter
Example
revenue week of quarter
week of year
Example
new products week of year last 3 years yearly

All of these keywords sort the data using datetime semantics, that is chronologically in a time sequence. By default, the Search bar suggests these keywords less frequently than others.

You can use these new keywords in combination with the existing data keywords which are:

  • Detailed

  • Hourly

  • Daily

  • Weekly

  • Monthly

  • Quarterly

  • Yearly

Examples of time series analysis

Typically, when you search for answers about series data, the visualizations that answer your questions are line charts. These charts frequently but not always include a stack to indicate a period.

time example1

When you search for an aspect of data of time series, a typical response is a line chart showing the aspect as it rises and falls over time.

time example2

You can also add a relative date filter for example,

total revenue quarterly yearly by year month(commit date) >= 01/01/1995
month(commit date) before 01/01/1998

This type of query also yields a stacked line chart:

time example3

The child date time attribute is on the X-axis and the parent in the legend. For example, if you search revenue month yearly the child, monthly, appears on the x-axis and the parent, yearly, in the legend.

Granularity for date filters

You can refine a simple date filters by adding hierarchical date filter to your query. The ability to specify two bucket granularities such as "hour of day" or "week of year" are two examples. The syntax of this type of query is

small_bucket of big_bucket [INTEGER_CONDITION]

The INTEGER_CONDITION is optional but it must be an integer. For example, this query is valid:

revenue by day of week <= 2

This query is invalid:

revenue by day of week = Tuesday

You can specify one or more granular filters.

These tips and gotchas apply to time granularity:

  • The system-defined fiscal rules are respected. This means, for example, if the fiscal year begins in February, month of year = 2 matches dates in March.

  • Fiscal shorthands such as Q1, Q2 and so on are not supported, so day of week = d1 is not valid.

  • INTEGER_CONDITION with = or != accept a list of filter values, so, day of week = 1 2 3 is valid.

  • INTEGER_CONDITION with = or != require legal values, so day of week > accepts any integer on the right hand side while day of week = requires a value in the legal 1-7 range.

  • Simple date filters allow you to use edit the filter through the answer to refine your search, adding a a hierarchical date filter in the search bar disables this ability.

Create a max(date) field and use it to filter

If you have a date field in your data set and want to return the most recent set of data based on a specific date, do the following:

  1. Create a formula called Max Date.

    For example:

     date = group_max ( date_to_filter_by )
  2. In the search bar, filter your dates by this formula.

    For example:

     max date = true

    This returns only those fields that pass the filter.