pablo-source / targets-test

Using Targets to automate markdown reports as a pipeline including TBATS ARIMA forecast

targets-test

Project to practise creating analytical pipelines to run models using {targets} library.

• The {targets} R package user manual: https://books.ropensci.org/targets/
• Targets walkthrough: https://books.ropensci.org/targets/walkthrough.html

Important:

• Each pipeline has its unique "_targets.R" file. And each pipeline will contain specific set of tar_target() and tar_group_by() functions used to configure the pipeline structure for each project.

• As this _targets.R file must retain its original name, I will save each _targets.R file for each pipeline in a separate folder in this GitHub project.

• Each pipeline folder will have to be run on a dedicated and individual R project to match the targets list from _targets.R file for that pipeline with its related set of adhoc R functions stored in the \R folder

• This will ensure each pipeline works for the purpose stated in the pipeline folder created at the top of this project.

• So when downloading each pipeline folder, it will contain the "_targets.R" files and related functions saved in the \R folder. All required input files will be sourced from the \data folder

Pipeline_01_populate_markdown_with_targets files:

• _targets.R (Specific Pipeline setup file)
• populate_markdown_with_targets_functions.R (Specific functions to populate this pipeline)

Pipeline_02_to_render_markdown:

• _targets.R (Specific Pipeline setup file)
• pipeline_render_markdown_functions.R (Specific function to populate this pipeline)

Pipeline_03_dynamic_branching files:

• _targets.R (Specific Pipeline setup file)
• dynamic_pipeline_functions.R (Specific functions to populate this pipeline)

1. Targets quick start guide

After installing the package, we load targets “library(targets)”. Then our first step is to run “use_tergets()” function. This creates a new file called _tragets.R that is used to configure and setup the pipeline.

Follow these steps then detailed in the R Documentation section of the use_targets() function:

After you call use_targets(), there is still configuration left to do:

Open ⁠_targets.R⁠ and edit by hand. Follow the comments to write any options, packages, and target definitions that your pipeline requires.

Edit run.R and choose which pipeline function to execute (tar_make(), tar_make_clustermq(), or tar_make_future()).

If applicable, edit clustermq.tmpl and/or future.tmpl to configure settings for your resource manager.

If applicable, configure job.sh, "clustermq.tmpl", and/or "future.tmpl" for your resource manager.

1.1 Create single scripts for each analysis steps

• In this example I have started creating one script to load the data and another one to create a plot from that data

• See script: before_targets/code_pre_targets.R

1.2 Turn these single scripts into functions

There is a folder called "before targets" containing individual R scripts called "code_pre_targets.R" this script allows me to plan the analysis. The second script "scripts_into_functions_targets_prep.R" contains new functions based on initial scripts to wwork with Targets package

• See script: before_targets/scripts_to_functions.R

1.3 Functions used by Targets saved in R folder

• The set of functions we want to run as part of our pipeline, are saved in the R folder for Targets to use them when executing the pipeline

• see script "study_functions.R" initial scripts for each analysis step turned into functions to be used in targets pipeline

• See script: R/study_functions.R

1.4 Pipeline defined in the _targets.R file

pipeline

• 1-4 Read in data
• 2-4 Clean data
• 3-4 Merge files
• 3-4 Plot data
• After the pipeline run we can run report.Rmd Markdown report and populate it with objects created by Targets pipeline. )

All required files to run this pipeline saved in folder: Pipeline_04_data_wrangling_union_merge

1.5 Specific {targets} functions used to execute the pipeline

Load targets library library(targets)

• First check for errors in the pipeline using tar_manifest() function tar_manifest(fields = command)

• Then check pipeline dependency graph using tar_visnetwork() function tar_visnetwork()

• Finally we run the pipeline we just built earlier using tar_make() function tar_make()

The plot created from our pipeline is now saved as an individual .png chart

1.6 Run pipeline

Fnally we run the pipeline we just built earlier using tar_make() function This function runs the correct targets in the correct order and saves the results to files tar_make()

Pipeline 01. populate markdown with targets

Everytime we update something in the pipeline we use "tar_make()" to re-run the entire pipeline. If some of the targets have not changed since last time we ran the pipeline, targets will skip those nodes in the pipeline called targets.

The tar_read() function we collect the pipeline output object to be used in specific sections of the Markdown report. For example, to use the data frame we creaetd on the first target we use tar_read(data). To use in the Markdown report the plot we created in the second Target object we use tar_read(plot). This allows us to populate our markdown report with specific objects created alongside the pipeline we just built and ran.

The final output of this pipeline is being used to create a fully rendered markdown report produced by the markdown file report.Rmd has been created and published in this repo:

The last step of this project has been building and rendering a markdown report called report.Rmd populated with the objects created in the pipeline by Targets. The aim is to autonmate the reports creation tasks by running a pipeline making it easier to mantain and update this report in the future. When rendering report.Rmd we obtain a document populated with tables and content from the pipeline. This could be expanded to automate reports ensuring reproducibility. Trying to follow RAP principles.

So now we have an initial pipeline that we can start to modify and expand to include extra analytical steps in the form of new targets

Pipeline 01. General pipeline structure using visnetwork

First we will merge all incoming .csv files, then we combine them into a single file and we use this new combined data frmae to populate our Markdown report.

This is the output usuing tar_visnetwork() function to check pipeline dependency graph

As part of the data preparation stage for future modelling pipeline

Pipeline 01. Completed pipeline final output

This is the output of the complated pipeline run, with dataframes saved and required .csv files saved in the \objects folder

After using tar_make() function we get the complete report of which sections of the pipeline have ran

All required files to run this pipeline saved in folder: Pipeline_01_populate_markdown_with_targets

1.Pipeline 02. Render Markdown in pipeline

We can render a Markdown document in the Targets pipeline by using {tarchetypes} library. This library provide us with the tar_render() function. So by adding a new target to our pipeline, we can render the report after the pipeline has run and it has populated our Markdown report.

And the rendering Targets function is now included in the pipeline:

After running the _targets file from this folder, we can automate the creation and rendering of a Markdown document inside the Targets pipeline

All required files to run this pipeline saved in folder: Pipeline_02_to_render_markdown

2.Pipeline 03. Dynamic branching and Time Series models forecast

Once the pipeline has run, before we implement a new feature (including a simple ARIMA model) defined in issue '#6', I have run fs:dir_tree("targets-test") to check whole set of objects created by Targets. The Markdown report has been populated by the three plots created in the pipeline.

In the coming week, I will be using Dynamic branching alongside Modeltime packages to introduce a couple of predictive models (ARIMA,Prophet) in the eixisting Pipeline. This is aimed to predict the next 5 months of Manufacturer's Value of Shipment for the following set of Shipment categories described below:

2.1 Dynanic branching

It is a way to define new targets while the pipeline is running. Opposed to declaring several targets up front. It is when you want to iterate over what is in the data, and you want a target that iterates by region. -Dynamic branching using {targets} https://books.ropensci.org/targets/dynamic.html

I will be using Dynamic branching to iterate over these four Economic Indicators downloaded from the FRED, Federal Reserve Economic Data:

Categories > Production & Business Activity > Manufacturing https://fred.stlouisfed.org/

Monthly time series indicators downloaded from FRED Economic Data. St Louis:

• Manufacturers' Value of Shipments: Total Manufacturing (AMTMVS). 2000-2024. Frequency: Monthly. U.S. Census Bureau, Manufacturers' Value of Shipments: Total Manufacturing [AMTMVS], retrieved from FRED, Federal Reserve Bank of St. Louis; April 2, 2024.Frequency: Monthly. Units: Millions of Dollars, Seasonally Adjusted. URL: https://fred.stlouisfed.org/series/AMTMVS
• Manufacturers' Value of Shipments: Computers and Electronic Products (A34SVS). 2000-2024. U.S. Census Bureau, Manufacturers' Value of Shipments: Computers and Electronic Products [A34SVS], retrieved from FRED, Federal Reserve Bank of St. Louis; April 3, 2024. Frequency: Monthly. Units: Millions of Dollars, Seasonally Adjusted. URL: https://fred.stlouisfed.org/series/A34SVS
• Manufacturers' Value of Shipments: Durable Goods (AMDMVS). 2000-2024. Frequency: Monthly U.S. Census Bureau, Manufacturers' Value of Shipments: Durable Goods [AMDMVS], retrieved from FRED, Federal Reserve Bank of St. Louis; April 2, 2024.Frequency: Monthly. Units: Millions of Dollars, Seasonally Adjusted. URL: https://fred.stlouisfed.org/series/AMDMVS
• Manufacturers' Value of Shipments: Nondefense Capital Goods Excluding Aircraft (ANXAVS). 2000-2024. Frequency: Monthly U.S. Census Bureau, Manufacturers' Value of Shipments: Nondefense Capital Goods Excluding Aircraft [ANXAVS], retrieved from FRED, Federal Reserve Bank of St. Louis; April 2, 2024.Frequency: Monthly. Units: Millions of Dollars, Seasonally Adjusted. URL: https://fred.stlouisfed.org/series/ANXAVS

This is an example of dynamic branching using tarchetypes package based on Metric variable, creating 2 branches for the two metrics included in this workflow: tarchetypes package GitHub repo:https://github.com/ropensci/tarchetypes/tree/main

Visnetwork from the above workfow including branching

All required files to run this pipeline saved in folder: Pipeline_03_dynamic_branching_files

• This will allow me using Modeltime, to apply each model to the different branches created by Targets, so the model will ran by each metric in the pipeline

4.Pipeline 05. Dynamic branching including ARIMA and Prophet models

This pipeline is completed and all required files to run it can be found in "Pipeline_05_ARIMA_Prophet_models" folder:

• Specific files to replicate this pipeline: "_targets.R","using_dynamic_predictive_pipeline.R" and "dynamic_predictive_pipeline_functions.R" this last one must be run from the \R folder

Using Modeltime Package to combine Prophet and ARIMA models in the previous Targets Pipeline. Modeltime package: https://business-science.github.io/modeltime/

targets-test's Issues

Update pipeline to get input files from /data sub folder

In the pipeline the second target "Clean data" now takes the input from previous target called now "file_csv"

Include an ARIMA model after the data_prep_models target.

After running the pipeline, data is in the right format to start building univariate TS models out of the input data. This issue is about including an ARIMA model in the pipeline

Update TARGETS package, combining the two scripts I created on my work laptop

Perform these set of actions in a Targets Pipeline:

• To tidy up existing Targets example

  • Merge three .csv files: AE_TYPE_1_ATT_AUG10_JAN24.csv,AE_TYPE_2_ATT_AUG10_JAN24.csv, AE_TYPE_3_ATT_AUG10_JAN24.csv
  • Create a GGPLOT chart using merged file
  • Save ggplot2 output into the objects folder
  • Force targets to remove some targets so next time we run tar_make() all pipeline will be re-build, so the output (.csv files)we produce in those targets will be re-created again.
  • tar_invalidate(one_two_combined)
  tar_invalidate(one_two_three_combined)

Create using Dynamic branching two TS models ARIMA and PROPHET

First I will combine the following four files for three new TS economic indicators below in a new pipeline to run separate TS models on each of them.

A34SVS_Shipments_value_Computer_products.csv
AMDMVS_Shipments_value_Durable_goods.csv
AMTMVS_Shipments_Total_Manufacturing.csv
ANXAVS_Shipments_value_Nondefense_Capital_Goods.csv

This time I will be using Dynamic branching from {targets} alongside {timetk} and {modeltime} packages to automate the creation of different models on different branches of the pipeline

Render markdown report from pipeline including text with inline figures

Once we have created the pipeline, we need to include all three plots created in the pipeline for TypeI, TypeII, and TypeIII attendances into the markdown report.

Also, create one paragraph for each Attendances type section where I populate it with figures and calculations from the dataset created as a result of running the pipeline, describing latest attendances figures for the latest complete year of data in the original ingested dataset.

Include data folder to store input files for pipeline

Create new folder named "data" where I will store any input .csv or xlsx file used in the pipeline

Include three new charts in Markdown report

Create three new line charts on the Targets pipeline to populate Markdown report, each one for them for Type_1_Attendances, Type_2_Attendances and Type_3_Attendances over time .

Then update existing Markdown report with these new three plots

Combine input files into single file using pipeline and join_by in DPLYR

Combine three .csv files AETypeI attendances, AEType2 attendances, AEType2 attendances downloaded from NHS England website into a single file by creating several pipeline functions in Targets using DPLYR

Include new ARIMA and PROPHET models after data_prep_models target

This will be the second uni variate TS model included in the pipeline

Populate Markdown report using Targets pipeline objects

The aim of this issue is to populate a Markdown report with objects (datasets and plots) produced by the targets pipeline created for this project.

• I will use A&E Attendances data from NHS England website to build both the pipeline and Markdown report

We start by loading required A&E Attendances Type I, Type II and Type III data from NHS England website. Link to the data:. Downloaded file "Monthly A&E Time Series March 2024 (XLS,403K)"https://www.england.nhs.uk/statistics/wp-content/uploads/sites/2/2024/04/Monthly-AE-Time-Series-March-2024.xls

And I will format this downloaded file to build a Markdown report, processing the required data for it using Targets pipeline.