Accumulating Multiple Variables with purrr::accumulate()

In the previous sections, we explored using purrr::accumulate() to create a custom function that predicts a variable based on its previous value. In this article, we will dive deeper into how to modify the function to accumulate two variables instead of just one.

Understanding the Problem

The original example used a simple model where the current prediction was dependent only on the lagged cumulative price (lag_cumprice) of the target variable. However, in some cases, it may be necessary to consider additional factors that affect the target variable’s behavior. In our variation model, we now have two lagged variables: cumrn (the cumulative row number) and lag_cumrn.

Modifying the Function

To modify the function accPrice() to accumulate both lag_cumprice and lag_cumrn, we need to introduce a new argument into the function. This new argument will hold the corresponding coefficient for the second lagged variable.

# Define the function with an additional argument
accPrice2 <- function(mod, acc1, acc2, cur) {
  
  # Get the grouped data segment
  db = cur_data_all()
  
  # Extract the current value of x (this iteration's input)
  x = db$x[cur]
  
  # Calculate the total exponent using all three terms
  total_exponent <- mod$coefficients['(Intercept)'] + 
    (mod$coefficients['x'] * x) +
    (mod$coefficients['lag_cumprice'] * acc1) + 
    (mod$coefficients['lag_cumrn'] * acc2)
  
  # Return the total exponent
  return(total_exponent)
}

# Now, we need to pass both accumulated values in the accumulate function
my_diamonds2 <- my_diamonds2 %>% 
  mutate(
    predicted = accumulate(.x = row_number()[-1], .init = InitialValue %>%
                          unique, .f = accPrice2, mod = silly_model)
  )

Key Takeaways

• To accommodate two lagged variables in the purrr::accumulate() function, we must pass both accumulated values as arguments to the custom function.
• The new argument should be assigned to a separate variable within the function (e.g., acc2) and used along with the existing accumulator (acc1).
• By using all three terms in the total exponent calculation, our model can now account for the effects of both lagged variables.

Considerations

The modification made here only allows us to accumulate two variables. If you need to accommodate more than that, you may have to consider more complex models or even additional techniques like using a different type of function (e.g., purrr::reduce()) or incorporating external libraries.

Example Code

To simplify the calculation and demonstrate how the new logic works, here is an example code snippet with some added comments:

# Example Data
d <- data.frame(cut = c("NA", "Int", "Premium"),
                x = c(100, 200, 300),
                cumprice = c(10, 15, 20),
                cumrn = c(1, 2, 3))

# Define the function with an additional argument
accPrice2 <- function(mod, acc1, acc2, cur) {
  # Get the grouped data segment
  db = d
  
  # Extract the current value of x (this iteration's input)
  x = db$x[cur]
  
  # Calculate the total exponent using all three terms
  total_exponent <- mod$coefficients['(Intercept)'] + 
    (mod$coefficients['x'] * x) +
    (mod$coefficients['lag_cumprice'] * acc1) + 
    (mod$coefficients['lag_cumrn'] * acc2)
  
  # Return the total exponent
  return(total_exponent)
}

# Define the model
model <- lm(cumprice ~ x + lag_cumprice + lag_cumrn, data = d)

# Fit the model
fit <- fit(model, init = c(0, 0, 1))

# Now, we need to pass both accumulated values in the accumulate function
d$predicted <- apply(fit, 1, accPrice2)

This is a basic demonstration of how you can modify your accPrice function to accommodate multiple lagged variables. By following these steps and adjusting your model as needed, you should be able to implement this approach in your own R projects.

Last modified on 2025-01-25