vjcitn/YESCDS
Cancer data science activities for CZI YES for CURE collaboration

Package index
Search the vjcitn/YESCDS package
Vignettes
Functions
82
Source code
29
Man pages
46
Home
/
GitHub
/
vjcitn/YESCDS
/

In vjcitn/YESCDS: Cancer data science activities for CZI YES for CURE collaboration

Elegant nonsense

With a very simple model of English grammar and a little vocabulary, we can randomly generate 'normal-sounding' sentences.

Here is a vector of elegant nouns.

nouns = c("Mercedes", "Tiffany", "Lamborghini", "Maserati", "Tosca", "La Traviata",
   "Diamonds", "Crystal", "Lenox")

We'll continue to set up some vectors of verbs (past tense), prepositions, adjectives, prepositions, and indicatives.

pastverbs = c("radiated", "glowed", "luxuriated", "ecstatized", "overwhelmed",
  "hyperventilated", "raptured", "raved", "stupendousized")
preps = c("on", "over", "beyond", "above", "beneath", "within", "across")
adjectives = c("magnificent", "majestic", "sophisticated", "exquisite",
  "glorious", "regal")
preps = c("on", "over", "beyond", "above", "beneath", "within", "across")
indic = c("the", "a", "any", "some")

Our model for a sentence is

(*) indicative + noun + verb(tensed) + preposition + indicative + adjective + noun

We produce a sentence by sampling purely randomly from the relevant vocabulary vectors. We use sprintf to substitute our choices into a template of seven strings.

get1 = function (x) sample(x, size=1)
up1 = function(x) paste0(toupper(substring(x,1,1)), substring(x,2))
sentence = function() {
  cat(sprintf("%s %s %s %s %s %s %s.", up1(get1(indic)), get1(nouns), get1(pastverbs),
     get1(preps), get1(indic), get1(adjectives), get1(nouns)))
}
set.seed(12345) # reproducible
sentence()
sentence()
sentence()
sentence()

The potential repetitiousness of purely random sampling can be avoided. In the following sentence generator, we prevent reuse of the initial noun or indicative.

sentence_nr = function() {
  i1 = get1(indic)
  n1 = get1(nouns) 
  pv1 = get1(pastverbs)
  p1 = get1(preps)
  i2 = get1(setdiff(indic,i1))   # don't repeat indicative
  ad = get1(adjectives) 
  n2 = get1(setdiff(nouns,n1)) # don't repeat noun
  sprintf("%s %s %s %s %s %s %s.", up1(i1), n1, pv1, p1, i2, ad, n2)
}
sentence_nr()
sentence_nr()

We can also compose longer sentences, using the conjunction "because":

linked = function() cat(sprintf("%s,\n because %s", sub(".$", "", sentence_nr()), sentence_nr()))
linked()
linked()

Exercise 1. What does the up1 function do? Apply it to the string "foo". What is the result?

Exercise 2. There is something wrong with the last two productions. What is it and how would you go about correcting it?

Exercise 3. Propose an alternate schematic to the one noted above as (*). Implement it in a function altsentence and illustrate its use with our vocaularies. For example 

noun + "is not" + noun

could be implemented as

altsentence = function() {
   sprintf("%s is not %s", get1(nouns), get1(nouns))
}
altsentence()
altsentence()

Come up with something more interesting than this!

Lexical tagging of English sentences

I asked perplexity.ai ($20/mo) to produce 10 lexically tagged sentences. By "lexical tagging" we refer to a process of associating with each word or phrase a category defining the word's role in sentence meaning. Examples of tags include

- DT: Determiner
- NN: Noun, singular or mass
- NNP: Proper noun, singular
- VB: Verb, base form
- VBZ: Verb, 3rd person singular present
- JJ: Adjective
- IN: Preposition or subordinating conjunction
- CD: Cardinal number

The full set of "part of speech" tags can be found at the U Penn linguistics site.

Here's what came back from perplexity.ai:

Statistics of parts of speech

The sentnce data do not possess a regular structure amenable to tabulation. There is a hierarchical format: arbitrary numbers of language particles and annotations are ordered within each sentence, and the sentences form a group of ten.

A popular format for irregularly structured data is "JSON". We asked perplexity.ai to produce the tagged sentences in JSON format. R can ingest JSON for programmatic analysis using the jsonlite package.

library(jsonlite)
myj = system.file("json/tensent.json", package="YESCDS")
sdat = fromJSON(myj, flatten=FALSE)
frames = sdat[[2]]
frames[[1]]

We see that for each sentence we have a data.frame with one row for each English word and associated part of speech.

The overall frequency distribution of parts of speech in this little collection is:

alltags = lapply(seq_len(length(frames)), function(x) frames[[x]][,2])
sort(table(unlist(alltags)))

Here's how we might use R to determine how many words into a sentence we need to go to find a verb:

sapply(frames, function(z) min(grep("VB", z[,2])))

Exercise 1. What is the position of the verb in the fourth sentence? What is the verb, and what type is it? Consult the link to U Penn linguistics given above.

Exercise 2. Which sentence has the verb in the "latest" position, and why is the position number so high? Should the counting procedure be adjusted?

Using an LLM

In this section we will assume that you are working on a machine that is able to use software and data organized at huggingface.co, specifically a version of BERT.

We will 'source' a github gist that will work if python is present, and if it and the reticulate package have been configured to be able to find python packages torch and transformers.

This code chunk will produce a table of scores and completions for the prompt The capital of Spain is [MASK]..

devtools::source_gist("https://gist.github.com/vjcitn/0e1b0289eb1a219b98d60b5b863f8d62")

This turns out to be difficult to use on the 'free' galaxy instance. The software can be demonstrated at https://vjcitn.shinyapps.io/AIapp. Of interest:

The code in the gist can be modified to change the model, and the prompt can be changed to seek scored responses to different prompts.



vjcitn/YESCDS documentation built on Oct. 30, 2024, 1:18 p.m.

R Package Documentation

rdrr.io home R language documentation Run R code online

Browse R Packages

CRAN packages Bioconductor packages R-Forge packages GitHub packages

We want your feedback!

Note that we can't provide technical support on individual packages. You should contact the package authors for that.
GitHub issue tracker
ian@mutexlabs.com
Personal blog

 
Embedding an R snippet on your website

Add the following code to your website.

For more information on customizing the embed code, read Embedding Snippets.

Close