A few weeks ago, I introduced a forecasting API (Application Programming Interface). The application can be found here:

https://techtonique2.herokuapp.com/

So far, as of 2022-11-23, this API contains four methods for univariate time series forecasting (with prediction intervals):

  • mean a (not so naïve) benchmark method, whose prediction is the sample mean.
  • rw a (not so naïve) benchmark method, whose prediction is the last value of the input time series.
  • theta is the forecasting method described in [1] and [2], which won the M3 competition.
  • prophet is a popular model described in [3].

In this post, I’ll present two packages, one implemented in R and one in Python, which are designed for smoothing users’ interaction with the API. You can create similar high-level packages in other programming languages, by using this tool and this page.

Content

  • 0 - Install packages in R or Python
  • 1 - Create an account with create_account
  • 2 - Get a token for authentication using get_token
  • 3 - Requests for forecasts with get_forecast

0 - Install packages in R or Python:

  • In Python
pip install forecastingapi
  • In R
library(devtools)
devtools::install_github("Techtonique/forecastingapi/R-package")
library(forecastingAPI)

1 - Create an account with create_account:

  • In Python
import forecastingapi as fapi

res_create_account = fapi.create_account(username="user1@example.com", password="pwd") # choose a better password
print(res_create_account)
  • In R
forecastingAPI::create_account(username = "user2@example.com", password = "pwd") # choose a better password

2 - Get a token for authentication using get_token

  • In Python
token = fapi.get_token(username = "user1@example.com", password = "pwd")
print(token)
  • In R
token <- forecastingAPI::get_token(username = "user2@example.com", password = "pwd")

The token is valid for 5 minutes. After 5 minutes, it must be renewed, using get_token.

3 - Requests for forecasts with get_forecast:

  • In Python
path_to_file = '/Users/t/Documents/datasets/time_series/univariate/USAccDeaths.csv' # (examples:https://github.com/Techtonique/datasets/tree/main/time_series/univariate)
    
res_get_forecast = fapi.get_forecast(file=path_to_file, token=token)

print(res_get_forecast)

res_get_forecast2 = fapi.get_forecast(file=path_to_file, 
token=token, start_training = 2, n_training = 7, h = 4, level = 90)

print(res_get_forecast2)

res_get_forecast3 = fapi.get_forecast(file=path_to_file, 
token=token, date_formatting="ms",
start_training = 2, n_training = 7, h = 4, level = 90)

print(res_get_forecast3)

res_get_forecast4 = fapi.get_forecast(file=path_to_file, 
token=token, method = "prophet")

print(res_get_forecast4)
  • In R
path_to_file <- '/Users/t/Documents/datasets/time_series/univariate/USAccDeaths.csv' # (examples:https://github.com/Techtonique/datasets/tree/main/time_series/univariate)

f_theta <- forecastingAPI::get_forecast(file = path_to_file, token = token,
                                        method = "theta", h=10, level = 95)

f_mean <- forecastingAPI::get_forecast(file = path_to_file, token = token,
                                       method = "mean", h=10, level = 95)

f_rw <- forecastingAPI::get_forecast(file = path_to_file, token = token,
                                     method = "rw", h=10, level = 95)

f_prophet <- forecastingAPI::get_forecast(file = path_to_file, token = token,
                                          method = "prophet", h=10, level = 95)

api results plot

[1] Assimakopoulos, V., & Nikolopoulos, K. (2000). The theta model: a decomposition approach to forecasting. International journal of forecasting, 16(4), 521-530.

[2] Hyndman, R. J., & Billah, B. (2003). Unmasking the Theta method. International Journal of Forecasting, 19(2), 287-290.

[3] Taylor, S. J., & Letham, B. (2018). Forecasting at scale. The American Statistician, 72(1), 37-45.

Comments powered by Talkyard.