Hey there! I get that you're working on time series analysis and were told to use R's ts() function, but your data has irregular time points (0,5,11,15,20,30,50) with no fixed periodicity—totally makes sense that the standard tutorials (focused on monthly/quarterly/annual data) aren't helpful here. Let's break down how to handle this:

First, Understand the Limitation of ts() #

The ts() function in R is built primarily for equally spaced time series. Its frequency parameter is designed for fixed cycles (like 12 for monthly, 4 for quarterly), which doesn't apply to your irregular time points. That said, you can still use ts()—you just need to work around this by attaching your custom time points as an attribute.

Step-by-Step Solution with ts() #

Let's walk through a concrete example:

1. Prepare your data and time points
   Suppose you have a vector of observed values and your custom time points:

   # Replace these with your actual data
   observed_values <- c(12, 23, 18, 27, 31, 25, 19)
   custom_time_points <- c(0, 5, 11, 15, 20, 30, 50)
   

2. Create the base ts object
   Since there's no fixed period, set frequency = 1 (this is the default if you omit it, but being explicit helps):

   my_ts <- ts(observed_values, frequency = 1)
   

3. Attach your custom time points as an attribute
   This lets you link the ts object to your actual time points without breaking the ts structure:

   attr(my_ts, "custom_times") <- custom_time_points
   

4. Use and verify the object
   You can now access both the ts data and your custom time points:

   # View the ts object
   my_ts
   
   # Retrieve your custom time points
   attr(my_ts, "custom_times")
   
   # Combine into a readable data frame
   data.frame(Time = attr(my_ts, "custom_times"), Value = as.vector(my_ts))
   

A Better Alternative (If You Can Flex the Tool) #

While ts() works with the above workaround, packages like zoo or xts are purpose-built for irregularly spaced time series. They let you directly index your data with custom time points, which is more intuitive for this use case. Here's a quick example with zoo:

library(zoo)

# Create a zoo object directly from your values and time points
my_zoo <- zoo(observed_values, order.by = custom_time_points)

# View the object (it will show your actual time points)
my_zoo

Just keep in mind: if you're required to stick with ts() for specific workflows, the attribute method will let you comply while keeping your time point context intact. Just be aware that some functions that assume equally spaced data (like certain forecasting tools) might not work as expected with this setup.

内容的提问来源于stack exchange，提问作者Maria Faleeva