Dynamic Modeling and Machine Learning Environment • Dyn4cast

The Dyn4cast package is designed to be a lightweight package. The philosophy behind it was the need to provide quick updates and visualization of Nigerian COVID-19 cases during the pandemic in 2020, hence had only one function, DynamicForecast. The aim was to simplify the estimation, prediction and forecast of time-varying COVID-19 dataset, based on daily update of incidences. There was need to have a function which is able to handle continuous data collection, estimation and forecast. That was what led to the name of the package. From that single function, the number of functions have grown to more than 10 because there were need to have supporting functions to make the forecasting easy. However, the package has pride itself of the line line technology by providing various machine learning functions that have the functionality of revealing the facts behind your data. The functions are working optimally and efforts are continuously being made to improve on them and ensuring that dependencies are reduced to the barest minimum. The unique selling point of this package is that it takes away the need to load multiple libraries to perform the various machine learning tasks.

Installation

Although it would be possible to install the released version of Dyn4cast from CRAN in future, presently, only the development version is available. The canonical form for CRAN is:

install.packages("Dyn4cast")

The development version is the only one available now and can be installed from GitHub with:

# install.packages("devtools")
devtools::install_github("JobNmadu/Dyn4cast")

The development version can also be installed through r-universe. Use the form:

install.packages("Dyn4cast", repos = c("https://jobnmadu.r-universe.dev", "https://cloud.r-project.org"))

Basic usage

At present, the package exports the following functions:

DynamicForecast which takes two required arguments: the Data of any recognized format but should be a dataframe containing two columns Date and Case. The Date is the day/month/year the data is collected while Case is the variable for forecasting. The Date must be in the recognized format i.e. ‘YYYY-MM-DD’. The other arguments parsed to the function are MaximumDate, which is the last date Data was collected and BREAKS, which is a vector of numbers and is used as knots in estimating spline polynomials.
constrainedforecast which constrain forecast of one-sided integer forecast to lie between the lower and upper limits of the base data. The function estimates the lower and upper 80% and 95% forecasts of the estimated model. This function works with two other functions, that is, invscaledligit and scaledlogit which are adapted from Hyndman & Athanasopoulos (2021) and both of which are adopted.
Percent which affix the % sign on a value or a vector or data frame of values.
MLMetrics which collects more than 40 metrics that are useful in model selection. The beauty of this function is the simplicity with which these metrics are collected from difference packages and saves the user the need to load more than 10 libraries in order to get these metrics.
MallowsCP for determining the the Mallows CP.
Linearsystems for linear regression model with some transformation.
quicksummary which outputs a formatted table of useful summary statistics of machine learning data.
formattedcut is a wrapper for providing publication ready frequency tables for continuous variable.
data_transform is a wrapper for standardizing data.frame to make the values comparable for estimation and/or visualization.
estimate_plot is a function for plotting estimated coefficients of a model in their order of significance.
corplot is for plotting the correlation matrix.
garrett_ranking is for ranking Likert-type data.
Model_factors is for determining and retrieving latent factors from Likert-type data for estimation and Machine Learning.
treatment_model is for propensity matching treatments effects and other metrics in the Machine Learning Environment.

Things the package can do

The package is capable of

computing, estimating, predicting and forecasting of the following models.
- Spline without knots
- Spline with knots
- Smooth Spline
- ARIMA
- Quadratic
- Ensembled with equal weight
- Ensembled based on weight
- Ensembled based on summed weight
- Ensembled based on weight of fit
Unconstrained forecasts
Constrained forecast
Machine Learning Metrics
Mallow’s CP
Per cent sign
- Rate
- percent
Scaled logit for constrained forecast
Inverse scaled logit for constrained forecast
Linear regression and functional forms which consists of
- Linear model
- Linear model with interactions
- Semilog model
- Growth model
- Double Log model
- Mixed-power model
- Translog model
- Quadratic model
- Cubic model
- Inverse of y model
- Inverse of x model
- Inverse of y & x model
- Square root model
- Cubic root model
- formatted Model Table
- Prediction plots
- Fitted plots
- Naive effects plots
- Summary of numeric variables
- Summary of character variables
Convert a continuous vector to a data frame
Convert a raw data frame to a uniform data frame
Plot of correlation matrix
Plot of the order of significance of estimates coefficients
Rank Likert-type data using Garrett ranking technique
Determine and retrieve the latent factors in Likert-type variables
Treatment model which is for propensity matching and treatment effects. It has the capacity to provide:
- Estimated treatment effects model
- Data frame of the estimated various treatment effects
- Vector of estimated propensity scores from the model
- Vector of fitted values from the model
- Residuals of the estimated model
- Plot of the propensity scores from the model faceted into Treated and control populations
- Plot of the average treatment effect for the entire population
- Plot of the average treatment effect for the treated population
- Plot of the average treatment effect for the controlled population
- Plot of the average Treatment effect for the evenly population
- Plot of the average Treatment effect for the overlap population
- Estimated weights for each of the treatment effects

Citation

The citation information for this package can be obtained easily when you run citation("Dyn4cast") in your R console.


citation("Dyn4cast")
To cite package 'Dyn4cast' in publications use:

  Nmadu J (2024). _Dyn4cast: Dynamic Modeling and Machine Learning
  Environment_. R package version 11.11.24, commit
  68dcb48ed8448692f4e25d5840387bee859a2add,
  <https://github.com/JobNmadu/Dyn4cast>.

A BibTeX entry for LaTeX users is

  @Manual{,
    title = {Dyn4cast: Dynamic Modeling and Machine Learning Environment},
    author = {Job Nmadu},
    year = {2024},
    note = {R package version 11.11.24, commit 68dcb48ed8448692f4e25d5840387bee859a2add},
    url = {https://github.com/JobNmadu/Dyn4cast},
  }

Suggested packages

Although not a dependency, the package derives functionally from a number of other packages and so may require you to install such packages if they are not yet installed. The packages are listed below:

install.packages(c("forecast", "lubridate", "Metrics", "tidyr", "ggplot2", "magrittr", "formattable", "xlsx", "readxl"))

Note that a warning (not error) is thrown up while estimating the RMSE for the Ensembled with equal weight model. It was thoroughly investigated and causes no harm. Efforts are still on to silence the warning, which I will soon. The warning is one of such issues that is general to R. If you set your chunk option to warning = FALSE you will not notice the warning.

Other suggestions?

The package is still very much in progress as such feedback, particularly at this developmental stage, would be greatly welcome and appreciated. Please fork your feedback at GitHub.

Bibliography

Mahoney, M. (2021). Model averaging methods: how and why to build ensemble models.

Hyndman, R. J. (2020). Quantile forecasting with ensembles and combinations in Business Forecasting: The Emerging Role of Artificial Intelligence and Machine Learning, eds. Gilliland, Tashman & Sglavo. John Wiley & Sons.

Hyndman, R.J., & Athanasopoulos, G. (2021). Forecasting: principles and practice, 3rd edition, OTexts: Melbourne, Australia. OTexts.com/fpp3. Accessed on July 30, 2021.

Lucy D’Agostino McGowan (2019). Understanding propensity score weighting. Available at: https://livefreeordichotomize.com/posts/2019-01-17-understanding-propensity-score-weighting/

Code of Conduct

Please note that the Dyn4cast project is released with a Contributor Code of Conduct. By contributing to this project, you agree to abide by its terms.