knitr::opts_chunk$set( message = FALSE, warning = FALSE, fig.width = 8, fig.height = 4.5, fig.align = 'center', out.width='95%', dpi = 100 ) # devtools::load_all() # Travis CI fails on load_all()
This vignette covers Machine Learning for Forecasting using the time-series signature, a collection calendar features derived from the timestamps in the time series.
The time series signature is a collection of useful features that describe the time series index of a time-based data set. It contains a wealth of features that can be used to forecast time series that contain patterns.
In this vignette, the user will learn methods to implement machine learning to predict future outcomes in a time-based data set. The vignette example uses a well known time series dataset, the Bike Sharing Dataset, from the UCI Machine Learning Repository. The vignette follows an example where we'll use timetk
to build a basic Machine Learning model to predict future values using the time series signature. The objective is to build a model and predict the next six months of Bike Sharing daily counts.
Before we get started, load the following packages.
library(dplyr) library(timetk) library(recipes) library(parsnip) library(workflows) library(rsample) # Used to convert plots from interactive to static interactive = FALSE
We'll be using the Bike Sharing Dataset from the UCI Machine Learning Repository.
Source: Fanaee-T, Hadi, and Gama, Joao, 'Event labeling combining ensemble detectors and background knowledge', Progress in Artificial Intelligence (2013): pp. 1-15, Springer Berlin Heidelberg
# Read data bike_transactions_tbl <- bike_sharing_daily %>% select(date = dteday, value = cnt) bike_transactions_tbl
Next, visualize the dataset with the plot_time_series()
function. Toggle .interactive = TRUE
to get a plotly interactive plot. FALSE
returns a ggplot2 static plot.
bike_transactions_tbl %>% plot_time_series(date, value, .interactive = interactive)
Next, use time_series_split()
to make a train/test set.
assess = "3 months"
tells the function to use the last 3-months of data as the testing set. cumulative = TRUE
tells the sampling to use all of the prior data as the training set. splits <- bike_transactions_tbl %>% time_series_split(assess = "3 months", cumulative = TRUE)
Next, visualize the train/test split.
tk_time_series_cv_plan()
: Converts the splits object to a data frame plot_time_series_cv_plan()
: Plots the time series sampling data using the "date" and "value" columns. splits %>% tk_time_series_cv_plan() %>% plot_time_series_cv_plan(date, value, .interactive = interactive)
Machine learning models are more complex than univariate models (e.g. ARIMA, Exponential Smoothing). This complexity typically requires a workflow (sometimes called a pipeline in other languages). The general process goes like this:
The first step is to add the time series signature to the training set, which will be used this to learn the patterns. New in timetk
0.1.3 is integration with the recipes
R package:
The recipes
package allows us to add preprocessing steps that are applied sequentially as part of a data transformation pipeline.
The timetk
has step_timeseries_signature()
, which is used to add a number of features that can help machine learning models.
library(recipes) # Add time series signature recipe_spec_timeseries <- recipe(value ~ ., data = training(splits)) %>% step_timeseries_signature(date)
We can see what happens when we apply a prepared recipe prep()
using the bake()
function. Many new columns were added from the timestamp "date" feature. These are features we can use in our machine learning models.
bake(prep(recipe_spec_timeseries), new_data = training(splits))
Next, I apply various preprocessing steps to improve the modeling behavior. If you wish to learn more, I have an Advanced Time Series course that will help you learn these techniques.
recipe_spec_final <- recipe_spec_timeseries %>% step_fourier(date, period = 365, K = 5) %>% step_rm(date) %>% step_rm(contains("iso"), contains("minute"), contains("hour"), contains("am.pm"), contains("xts")) %>% step_normalize(contains("index.num"), date_year) %>% step_dummy(contains("lbl"), one_hot = TRUE) juice(prep(recipe_spec_final))
Next, let's create a model specification. We'll use a Elastic Net penalized regression via the glmnet
package.
model_spec_lm <- linear_reg( mode = "regression", penalty = 0.1 ) %>% set_engine("glmnet")
We can mary up the preprocessing recipe and the model using a workflow()
.
workflow_lm <- workflow() %>% add_recipe(recipe_spec_final) %>% add_model(model_spec_lm) workflow_lm
The workflow can be trained with the fit()
function.
if (requireNamespace("glmnet")) { workflow_fit_lm <- workflow_lm %>% fit(data = training(splits)) }
Linear regression has no parameters. Therefore, this step is not needed. More complex models have hyperparameters that require tuning. Algorithms include:
If you would like to learn how to tune these models for time series, then join the waitlist for my advanced Time Series Analysis & Forecasting Course.
The Modeltime Workflow is designed to speed up model evaluation and selection. Now that we have several time series models, let's analyze them and forecast the future with the modeltime
package.
The Modeltime Table organizes the models with IDs and creates generic descriptions to help us keep track of our models. Let's add the models to a modeltime_table()
.
if (rlang::is_installed("modeltime")) { model_table <- modeltime::modeltime_table( workflow_fit_lm ) model_table }
Model Calibration is used to quantify error and estimate confidence intervals. We'll perform model calibration on the out-of-sample data (aka. the Testing Set) with the modeltime::modeltime_calibrate()
function. Two new columns are generated (".type" and ".calibration_data"), the most important of which is the ".calibration_data". This includes the actual values, fitted values, and residuals for the testing set.
calibration_table <- model_table %>% modeltime::modeltime_calibrate(testing(splits)) calibration_table
With calibrated data, we can visualize the testing predictions (forecast).
modeltime::modeltime_forecast()
to generate the forecast data for the testing set as a tibble. modeltime::plot_modeltime_forecast()
to visualize the results in interactive and static plot formats.calibration_table %>% modeltime::modeltime_forecast(actual_data = bike_transactions_tbl) %>% modeltime::plot_modeltime_forecast(.interactive = interactive)
Next, calculate the testing accuracy to compare the models.
modeltime::modeltime_accuracy()
to generate the out-of-sample accuracy metrics as a tibble.modeltime::table_modeltime_accuracy()
to generate interactive and static calibration_table %>% modeltime::modeltime_accuracy() %>% modeltime::table_modeltime_accuracy(.interactive = interactive)
Refitting is a best-practice before forecasting the future.
modeltime::modeltime_refit()
: We re-train on full data (bike_transactions_tbl
)modeltime::modeltime_forecast()
: For models that only depend on the "date" feature, we can use h
(horizon) to forecast forward. Setting h = "12 months"
forecasts then next 12-months of data. calibration_table %>% modeltime::modeltime_refit(bike_transactions_tbl) %>% modeltime::modeltime_forecast(h = "12 months", actual_data = bike_transactions_tbl) %>% modeltime::plot_modeltime_forecast(.interactive = interactive)
Timetk is part of the amazing Modeltime Ecosystem for time series forecasting. But it can take a long time to learn:
Your probably thinking how am I ever going to learn time series forecasting. Here's the solution that will save you years of struggling.
Become the forecasting expert for your organization
High-Performance Time Series Course
Time series is changing. Businesses now need 10,000+ time series forecasts every day. This is what I call a High-Performance Time Series Forecasting System (HPTSF) - Accurate, Robust, and Scalable Forecasting.
High-Performance Forecasting Systems will save companies by improving accuracy and scalability. Imagine what will happen to your career if you can provide your organization a "High-Performance Time Series Forecasting System" (HPTSF System).
I teach how to build a HPTFS System in my High-Performance Time Series Forecasting Course. You will learn:
Modeltime
- 30+ Models (Prophet, ARIMA, XGBoost, Random Forest, & many more)GluonTS
(Competition Winners)Become the Time Series Expert for your organization.
Take the High-Performance Time Series Forecasting Course
Any scripts or data that you put into this service are public.
Add the following code to your website.
For more information on customizing the embed code, read Embedding Snippets.