README.Rmd

---
title: "Projects for Text Version Comparison and Analytics in R"
output:
  html_document:
    keep_md: yes
    number_sections: yes
    self_contained: yes
    toc: yes
  pdf_document:
    number_sections: yes
    toc: yes
---




```{r setting options, include=FALSE}
library(knitr)

options("width"=101)
tmp <- devtools:::read_dcf("DESCRIPTION")
knitr::opts_chunk$set(message=FALSE)
```

```{r title, results='asis', echo=FALSE, include=FALSE}
cat("#", tmp$Title)
```

# Overview

**Status**

```{r, echo=FALSE, results='asis'}
doc_type <- opts_knit$get("rmarkdown.pandoc.to")
if( !is.null(doc_type) && doc_type=="html" ){
cat(
'
[![Travis-CI Build Status](https://travis-ci.org/petermeissner/diffrprojects.svg?branch=master)](https://travis-ci.org/petermeissner/diffrprojects)
[![Codecov test coverage](https://codecov.io/gh/petermeissner/diffrprojects/branch/master/graph/badge.svg)](https://codecov.io/gh/petermeissner/diffrprojects?branch=master)
[![CRAN version](http://www.r-pkg.org/badges/version/diffrprojects)](https://cran.r-project.org/package=diffrprojects)
'
)
}
```



```{r count lines of code, include=FALSE}
filelist.R   <- list.files("R", recursive = TRUE, pattern="\\.R$", ignore.case = TRUE, full.names = TRUE)
filelist.tests   <- list.files("tests", recursive = TRUE, pattern="\\.R$", ignore.case = TRUE, full.names = TRUE)
filelist.cpp <- list.files("src", recursive = TRUE, pattern="\\.cpp$", ignore.case = TRUE, full.names = TRUE)
lines.R      <- unlist(lapply(filelist.R, readLines))
lines.tests  <- unlist(lapply(filelist.tests, readLines))
lines.cpp    <- unlist(lapply(filelist.cpp, readLines))
length.R     <- 
  length(grep("(^\\s*$)|(^\\s*#)|(^\\s*//)", lines.R,  value = TRUE, invert = TRUE))
length.tests <- 
  length(grep("(^\\s*$)|(^\\s*#)|(^\\s*//)", lines.tests,  value = TRUE, invert = TRUE))
length.cpp   <- 
  length(grep("(^\\s*$)|(^\\s*#)|(^\\s*//)", lines.cpp,  value = TRUE, invert = TRUE))
```





*R code:* `r length.R`<br> 
*C++ code:* `r length.cpp`<br> 
*test code:* `r length.tests`





**Version**

```{r version, results='asis', echo=FALSE}
cat(tmp$Version)
```





**Description**

```{r description, results='asis', echo=FALSE}
cat(tmp$Description)
```




**Funding**

This software was created as part of the "Institutional Design in Western European Democracies" research project, funded by [DFG (Deutsche Forschungsgemeinschaft)](http://gepris.dfg.de/gepris/projekt/146229116), lead by [Ulrich Sieberer](https://scholar.google.com/citations?user=Q_ImhbkAAAAJ) and based at [University Konstanz](https://www.uni-konstanz.de/). 




**License**

```{r licence, results='asis', echo=FALSE}
cat(tmp$License, "<br>")
cat(as.character(eval(parse(text=tmp$Author))))
```




**Citation**


```{r citation1, results='asis', echo=FALSE}
citation("diffrprojects")
```

**BibTex for citing**

```{r citation2, eval=FALSE}
toBibtex(citation("diffrprojects"))
```




**Installation**

stable CRAN version

```{r installation1, eval=FALSE}
install.packages("diffrprojects")
library(rtext)
```



(stable) development version

```{r installation2, eval=FALSE}
standard_repos <- options("repos")$repos
install.packages( 
  "diffrprojects", 
  repos = c(standard_repos, "https://petermeissner.github.io/drat/")
)
library(rtext)
```




**Contribution**

Note, that this package uses a Contributor Code of Conduct. By participating in this project you agree to abide by its terms: http://contributor-covenant.org/version/1/0/0/ (basically this should be a place where people get along with each other respectfully and nicely, because it's simply more fun that way for everybody)

Contributions are very much welcome, e.g. in the form of:

- **typo fixing** ([edit file directly on Github](https://help.github.com/articles/editing-files-in-another-user-s-repository/))
- **bug reporting** (file an [issue](https://guides.github.com/features/issues/) - after having searched if the issue came up before - as - if possible - [minimal reproducable example](http://stackoverflow.com/help/mcve))
- **extending help files** (e.g. [edit the respective files directly on Github](https://help.github.com/articles/editing-files-in-another-user-s-repository/) or [fork the package](https://help.github.com/articles/fork-a-repo/) and later on make a [pull request](https://help.github.com/articles/using-pull-requests/); note, that the package use [roxygen2](http://r-pkgs.had.co.nz/man.html) for easing documentation)
- **writing example** (e.g. [edit the respective files directly on Github](https://help.github.com/articles/editing-files-in-another-user-s-repository/) or [fork the package](https://help.github.com/articles/fork-a-repo/) and later on make a [pull request](https://help.github.com/articles/using-pull-requests/); note, that the package use [roxygen2](http://r-pkgs.had.co.nz/man.html) for easing documentation)
- **vignette writing** (file an [issue](https://guides.github.com/features/issues/) first so that we can discuss htings than [fork the package](https://help.github.com/articles/fork-a-repo/) and later on make a [pull request](https://help.github.com/articles/using-pull-requests/))
- **test writing** (have a look at the [test coverage](https://codecov.io/gh/petermeissner/stringb/tree/master/R) than [fork the package](https://help.github.com/articles/fork-a-repo/) and later on make a [pull request](https://help.github.com/articles/using-pull-requests/))
- **feature suggestions** (file an [issue](https://guides.github.com/features/issues/) describing the idea, why this is important, possible alternative solutions and an example)
- **general discussion** of approach and or implementation  (file an [issue](https://guides.github.com/features/issues/))
- implementation **improvements** (file an [issue](https://guides.github.com/features/issues/) naming whats to be improved, why and how)


   

# Usage






## Fast Introduction for the Impatient

For those in a hurry here is a very brief

```{r fast intro, cache=TRUE}
# loading package
library(diffrprojects)


# the first chapter of Robinson Crusoe from three different sources
rcs <- rtext:::testfile(pattern="rc.*ch1.txt", full.names = TRUE) 


# creating a new project
dp <- diffrproject$new()


# setting options
dp$options$verbose <- FALSE


# adding texts to the corpus
dp$text_add(text_file = rcs)
dp$text_data(1) %>% head(11)


# linking the files (which file should be compared to which)
dp$text_link()
dp$link %>% as.data.frame()


# calculating text alignments
dp$text_align(tokenizer=text_tokenize_words)
dp$alignment[[1]] %>% head(30)
```
   
   
   
   
   
   
   
   
## Creating a Diffrprojects Instance

To create a diffrproject we use the diffrproject creator object - it´s simply an object with a function that knows how to create a project. 

Creating a project looks like this:

```{r new project}
library(diffrprojects)
dp <- diffrproject$new()
```

Et violà - we created a first, for now empty, project that we will use throughout the tutorial. 








## Some Help Please 

To get a better idea about what this thing called *diffrproject* really is, you can consult its help page which gives a broad overview over its capabilities:

```{r getting help, eval=FALSE}
?diffrproject
```

Another way is to call the ls() method. This will present us with a data frame listing, all fields where data is stored and all the methods (aka object specific functions) of our diffrprojects instance. Those methods and fields located in *private* are not for the user to mess around with while non-private (*self* aka public) data fields can be read by the user and public methods can be triggered by the user to manipulate the data or retrieve data in a specific format. 

```{r ls method}
dp$ls()
```

The base R class() function furthermore reveals from which classes the diffrproject class inherits: 

```{r class}
class(dp)
```







## Adding Texts to Projects

Our diffrproject (`dp`) has one method called `text_add()` that allows to add texts to the project. Basically the method can be used in three different flavors: adding character vectors, adding texts stored on disk, or by adding rtext objects (see rtext package:  https://CRAN.R-project.org/package=rtext; rtext objects are the way individual texts are represented within diffrprojects). For each of these used cases there is one option: `text`, `text_file`, `rtext`; respectively. 

Below are shown examples using each of these methods:


**adding text files**
```{r text_add files}
test_file1 <- stringb:::test_file("rc_1_ch1.txt")
test_file2 <- stringb:::test_file("rc_2_ch1.txt")
dp$text_add(text_file = c(test_file1, test_file2) )
```


**adding rtext objects**
```{r text_add rtext}
test_file <- stringb:::test_file("rc_1_ch1.txt")
rt <- rtext$new( text_file = test_file)
dp$text_add(rtext = rt)
```


**adding character vectors**
```{r text_add character vector}
test_file1 <- stringb:::test_file("rc_1_ch1.txt")
test_file2 <- stringb:::test_file("rc_2_ch1.txt")
cv <- ""
cv[1] <- text_read(test_file1, NULL)
cv[2] <- text_read(test_file2, NULL)
dp$text_add(text = cv)
```

In the last case make sure to put each text in one separate line. Functions like 
readLines() or text_read() read in texts such that each line corresponds to one element in a character vector. With e.g. text_read()'s tokenize parameter to NULL the text will be read in as one long string. 











## Piping Methods 

Now is a good time to mention a feature of diffrprojects that comes in handy: All functions that do not explicitly extract data (those usually have some 'get' as part of their name) do return the object itself so that one can pipe together a series of method calls.

Consider the following example where we initiate a new diffrprojects instance and add two texts in just one pipe:

```{r piping}
dp <- 
  diffrproject$
  new()$
  text_add(text_version_1, name = "version1")$
  text_add(text_version_2, name = "version2")

length(dp$text)
```




## Getting Infos About Texts 

If we want to get some general overview about the texts gathered in our project, we can use the text_meta_data() method to do so. The method has no parameters and returns a data.frame with several variables informing us about its source, length, encoding used for storage, and its name. 

```{r text meta data}
dp$text_meta_data()
```



## Showing Text

If you want to have a look at your texts you may do so by using the text's own text_show methods. Per default this method only shows the first 500 characters, but it can be set to higher numbers as well. 

```{r}
dp$text$version1$text_show(length=1000)
dp$text$version2$text_show(length=1000)
```



## Getting And Setting Infos About the Project

Similar to the text_meta_data() method we can access the projects meta data via data fields meta and options. But contrary to the text_meta_data() method that gathers data from all the texts within the project and does not allow for manipulation of the data, the data fields allow reading and writing. 

First let us have a look and thereafter turn off the message notification service:

**getting data fields**
```{r options get}
dp$options
```

**setting data fields**
```{r options set}
dp$options$verbose <- FALSE
```

(note, ask is deprecated and only remains for compatibility reasons but has no function anymore)

Now it's time to have a look at the projects meta data. It tells us when the project was created, which path to use for SQLite exports, which path to use for saving data as in RData format and what is the projects id. The id is a hash of a time stamp as well as session information which should ensure uniqueness across space and time. 

All these values can be manipulated by the user to her liking. 

```{r meta}
dp$meta

dp$meta$file_path = "./diffrproject.RData"
```







## Deleting Texts 

Of cause we can not only add texts but delete them from the project as well. For this purpose there is the text_delete() method.

Let's just add two texts and delete one by providing its index number and the second by providing its name to the text_delete() method. 

```{r text deletion}
dp$text_add(text = "nonesense", "n1")
dp$text_add(text = "nonesense", "n2")

dp$text_delete(3)
dp$text_delete("n2")

length(dp$text)
names(dp$text)
```







## Defining Relationships Between Texts: Linking

The purpose of diffrprojects is to enable data collection on the difference of texts. Having filled a project with various texts, there are endless possibilities to form pairs of text for comparison and change measurement - where endless actually is equal to: $n^2-n$. 

Linking can be done via the text_link method which accepts either index numbers or text names for its from and to arguments (a third argument delete will delete a specified link if set to TRUE).

```{r}
dp$text_link(from = 1, to = 2)
dp$text_link(from = 1, to = 2, delete = TRUE)
```

If no arguments are specified, text_link will link the first text to the second, the third to the fourth, the fourth to the fifths and so on.

```{r}
dp$text_link()
```


To get an idea of what links are currently specified, we can directly access the link data field or/and ask R to transform the list found there into a data.frame. 

```{r}
dp$link

dp$link %>% as.data.frame()
```








## Aligning Texts and Measuring Change

At the heart of each diffrproject lies the text_align method. This method compares two texts and tries to align parts of one text with parts of the other text. The first two arguments (`t1` and `t2`) are for specifying which pair of texts to compare - if left as-is, all text pairs that are specified within the link data field will be aligned. 

Text parts are arbitrary character spans defined by the `tokenizer` argument. This argument expects a function splitting text into a token data.frame. If the tokenizer argument is left as-is, it will default to text_tokenize_lines function from the stringb package. 

Text tokens can be pre-processed before alignment. The `clean` argument allows to hand over a function tranforming a charactr vector of text tokens into their clean counterparts. 

The `ignore` arguments expects a function that is able to transform a character vector of tokens into a logical vector of same length, indicating which tokens to ignore throughout the alignment process and which to consider. 

The next argument - `distance` - specifies which distance metrics to use to calculate distances between strings. 

Since the text_align method basically is a wrapper around diff_align you can get more information via `?diff_align` and since again diff_align is a wrapper around stringdist from the stringdist package `?stringdist::stringdist` and also ``?stringdist::`stringdist-metrics` `` will provide further insights about possible metrics and how to use the rest of the arguments to text_align (these are passed through to stringdist).

Let's have an example using the Levenshtein distance to calculate distances between tokens (lines per default). Furthermore we allow the distance between two aligned tokens to be as large as 15. Tokens which do not find a partner below that distance are considered to have been deleted or respectively inserted. Tokens which find a partner with a non-zero distance which is not above the threshhold are considered changes - transformations of one token into the other. 

The following shows the resulting list of alignment data.frames. 

```{r}
dp$text_align(distance = "lv", maxDist = 15)

dp$alignment
```


To measure the change between those two texts we can e.g. aggregate the the distances by change type: 

```{r}
sum_up_changes <- function(x){
  x %>% 
    dplyr::group_by(type) %>% 
    dplyr::summarise(sum_of_change = sum(distance))
}

lapply( dp$alignment, sum_up_changes)
```





## Coding Texts

Now let us put some data into our diffrproject. 

The most basic method to do so is simply called text_code. Text_code takes up to five arguments (the first three are mandatory), where one specifies the text to be coded (`text`, either by index number or by name), how the variable to store the information is called (`x`), and the index number or a vector of those indicating which characters of the text should be coded. The last two parameters are optional and specify which value the variable should hold (`val`) and at which hierarchy level the coding is placed (`hl`, higher or equal hierarchy levels will overwrite existing codings of lower hierarchy level for the same text, character span, and variable). 

```{r}
dp$text_code(text = 1, x = "start", i=1:5, val = TRUE, hl = 0)
dp$text_code(text = "version2", x = "start", i=1:5, val = TRUE, hl = 0)
```

The text_code method is quite verbose and in most cases more suited to be accessed by a machine or algorithm than by a human. Therefore, there are three other methods to code text: text_code_regex, text_code_alignment_token, text_code_alignment_token_regex. 

The text_code_regex method allows to search for text patterns and code a whole pattern instead of assigning codes character by character - the `i` argument of text_code gets replaced by a `pattern` argument. The in addition further arguments can be passed to the pattern search functions via `...` - see e.g. `?grep` for possible further arguments and https://stat.ethz.ch/R-manual/R-devel/library/base/html/regex.html for a description of regular expressions in R. 

In this example we are searching for the word *"it"* in text 1 and code each instance. 

```{r}
dp$text_code_regex(text = 1, x = "it", pattern = "\\bit\\b", ignore.case=TRUE)
```

Another variant of coding text is by using alignment tokens. Having alignment data availible, this allows for selecting: link, alignment and text while the other arguments from above stay the same. 


```{r}
# having a look at alignment number 4
dp$alignment[[1]][4,]

# coding text connected by alignment number 4
dp$text_code_alignment_token(
  link        = 1, 
  alignment_i = 4, 
  text1       = TRUE,
  text2       = TRUE,
  x   = "token_coding",
  val = 4,
  hl  = 0
)
```







## Getting Text Codings

The most basic way to get text data is to use the text_data method. This method will go through all or only selected texts, gather all the data stored there and put it into a neat data.frame where name identifies the text from which the data comes per name, char informs us about the character that was coded, and i refers to the characters position within the text. All other variables hold the data we added during the examples above. 

```{r}
dp$text_data(text = 1) %>% head()
```





## Aggregating Text Codings

The usage of text_data has its merits but often one is more interested in text data aggregated to a specific level. The following three aggregation functions offer a solution to this problem: tokenize_text_data_lines, tokenize_text_data_words, and tokenize_text_data_regex. These three methods make use of the similiary named methods provided by the rtext package. 

One important thing to keep in mind is that using these methods implies aggregating several data values on character level into one data value at token level. Therefore there has to be some aggregation function to be involved. The default is to use the value that occurs most often on character level, if more than one distinct values occur more than once the first is choosen. 

The aggregation function can be changed to whatever function the user seems appropriate by passing it to `aggregate_function` - as long as it reduces a vector of values into a vector with only one value. 

The `join` argument allows to decide how text and data are joined into the resulting data.frame - left: all token, right: all data, full: token with or without data and data with or without token.

```{r}
dp$tokenize_text_data_lines(
  text = 1, 
  join = "right", 
  aggregate_function = 
    function(x){
      paste(x[1:3], collapse = ",")
    }
) 
```






## Text Coding Inheritence

Having aligned two texts via token pairs another functionality of diffrprojects becomes availible: text coding inheritance via no-change tokens. This means that text codings can get copied to those tokens they are aligned with, given that they are considered the same - i.e. the distance equals zero and the change type therefore is no-change. 

To show this feature we use the text_inherit method and we will start with a fresh example. A new project with two texts. The first text gets some codings, then they are aligned, and in a last step codings are transfered from one text to the other via the text_data_inherit method. 


```{r}
dp <- 
  diffrproject$new()$
  text_add(text_version_1)$
  text_add(text_version_2)$
  text_code_regex(
    text    = 1, 
    x       = "test1", 
    pattern = "This part.*?change",
    val = "inherited"
  )$
  text_code_regex(
    text    = 1, 
    x       = "test2", 
    pattern = "This part.*?change",
    val = "inherited"
  )

dp$tokenize_text_data_lines(1)


dp$
  text_link()$
  text_align()$
  text_data_inherit(
    link      = 1,
    direction = "forward"
  )

dp$tokenize_text_data_lines(2)
```



## Saving and Loading Projects

Diffrprojects also allow for storing and loading project to and from disk. 

```{r, eval=FALSE}
# save to file
dp$save(file = "dp_save.RData")

# remove object
rm(dp)

# create new object and load saved data into new object
dp <- diffrproject$new()
dp$load("dp_save.RData")
```

```{r}
dp$tokenize_text_data_lines(2)
```








# Diffrprojectswidget a Diffrprojects Extension

```{r, include=FALSE}
dpw <- require(diffrprojectswidget)
wbs <- require(webshot)
dpw <- dpw & wbs
```


```{r, eval=dpw}
library(diffrprojectswidget)
dp_table(dp, 1, height = 800, width = "100%")
```


```{r, eval=dpw}
library(diffrprojectswidget)
dp_vis(dp, 1, height = 300)
```






# Technicalities



## Naming Conventions and General Structure of Methods and Data

The methods and data fields of diffrprojects can be categorized into five realms - *cursive*: methods; (parentheses): private; rest: data: 

- **text**: everything related to individual texts starts with text 
    - text, *text_add*, *text_delete*, *text_align*, *text_code*, *text_code_alignment_token*, *text_code_alignment_token_regex*, *text_code_regex*, 
    - *text_data*, *text_data_inherit*, *tokenize_text_data_lines*, *tokenize_text_data_regex*, *tokenize_text_data_words*
    - text_meta_data
- **alignment**: everything that concerns the relation between two texts
    - alignment, *alignment_add*, *alignment_code*, *alignment_delete*, *alignemtn_data_full*, *alignment_data_set*
    - text_link, *link*
- **misc**:
    - meta, options, *load*, *save*, *export_sqlite*, *import_sqlite*, (*execute_load*), (*prepare_save*)
- **inherited from R6_rtext_extended**:
    - options, *message*, *warning*, (*hash*), (hashed), (hashes)
- **inherited from R6**:
    - *clone*, *initialize*



## Data formats 


### meta

Meta is a list with only a few items providing/storing general information for the whole project - i.e. time stamp the project was created, path to store data, path to export data, an project id.


### text

Text is a list of rtext instances. Each rtext instance stores text's actual text as data gathered on the text. 

The text_data method will return a data.frame containing all text data, while tokenize_text_data_xxx methods will aggregate text data to specific token levels: words, lines or user defined patterns. 


### link

Link is a list of links between texts. Link defines for which text combination alignments should be calculated. Each list item hold a from and to field which stores the names of texts to be aligned. The method to create links is text_link, it also allows to delete specific links. 

Link data can be transformed to one big data.frame via: as.data.frame function. 


### alignment 

Alignment is a list of data.frames. Each alignment list item stores which part of one text (character span) is connected to which part of another text (character span). 

The list of alignments can be transformed to one big data.frame via: as.data.frame function. 


### alignment_data

Alignment_data is a list of dataframes. 




## The Diffrprojects Universe

Diffrprojects has two other packages it relies heavily on and one package that adds further features. 


### Rtext

Rtext is a package providing a data structure and accompanying methods to handle texts / strings / characters as well as data bound to these texts / strings / characters. All string manipulations are based upon the stringb package. All diffrproject texts are actually rtext instances. Unfortunate you cannot yet manipulate rtext objects once they are part of a diffrproject and expect that data on the relation between texts (i.e. alignment and alignment_data) gets updated as well - hence manipulating texts might lead to inconsistencies in alignments and alignment_data. 

A strategy to implement such a feature would be to extend rtext in such a way that text manipulation methods would pass change information to a list of call back functions. Furthermore, diffrprojects need two methods that allow for handling shifts in the character sequences of texts. Those update methods can then be passed to rtext instances once they become part of a diffrproject. Then whenever e.g. some characters are deleted, alignments as well as alignment data touching these character spans get deleted as well and character span information for all other alignments get shifted by the appropriate amount. 

For those preferring a version using stringi/stringr - go ahead - since rtext and diffrprojects provide tests for all respectively for all vital parts and stringb copied the function naming scheme from stringr anyways, this should be a small matter. 



### Stringb

Stringb is a package providing convenience functions for string handling and manipulation using R's own regular expression engine. All string manipulations are based upon the stringb package. 

In addition stringb provides very flexible text tokenization functions that are very much in line with the needs of diffrprojects. 



### Diffrprojectswidget


This package enhances diffrprojects by providing HTMLwidgets for visualizing diffrproject data: as interactive table or as interactive graph.  

HTMLwidgets (see: http://www.htmlwidgets.org/) are a framework that allows for interactive, web technology based graphics that are furthermore easily integrate able into e.g. R-shiny (http://shiny.rstudio.com/) applications. 


## Two words or more about Objects / R6 / Classes / Instances

Diffrprojects is written in object oriented programming style because it seemed adequate to do so. Why? Because in OOP in comparison to functional programming one does more stuff like in-place-modification of data, data and its modifiers (methods) come in one big bundle, it´s easier to work on the current state of the object / to only allow consistent states of the object. Yeap everything here could have done with FP as well - please go ahead. 

The downside of using OOP in R is that what happens becomes much more intrasparent and harder to reason about - I am sorry for that. 


### Classes and Instances

Classes are object blueprints - a schema that describes how an object of this class should look like. Classes might be objects too, but they are not the objects they describe. To get an object instance of an object - a manifestion of the idea of the object described in the class - one has to explicitly translate execute the instructions led out in the class, e.g. via: `diffrproject$new()`, or `rtext$new()`


### R6

R6 is a package that provides a framework that makes it very, very easy to build objects in R that are more like things known from traditional all purpose programming languages like Java or C++.  



### Methods

Methods are exactly like functions, only that they are not floating around loosely in your global environment or elsewhere, but are bound to specific instances of an object. So there is not one text_add function that can be used with any diffrproject, but there is one specific text_add method for each instance of an diffrproject. This sounds quite strange, right? Why the duplication? Well, with that you can e.g. pass this method around, hand it over to a function that calls it or put it into another object maybe that than can decide to use it or not. A silly example: 


```{r}
dp1 <- diffrproject$new()
add_text_to_dp1 <- dp1$text_add

add_text_to_dp1("ahhh")
add_text_to_dp1("behhh")
add_text_to_dp1("cehhh")

names(dp1$text)
```