use matra::domain::Format;
use matra::nlp::udpipe::Udpipe;
use matra::{Engine, Ingest};
fn main() {
let nlp = Udpipe::english("/tmp/matra-models").expect("Failed to load English model");
let engine = Engine::new(Box::new(nlp), matra::standard_decomposers());
let text = r#"
## The Problem
In 2016 I was part of a team assigned the mandate for bot detection.
We needed a programmable reverse proxy, and we picked Styx: JVM-based,
reactive, open source but built in-house. Envoy had been open sourced
that year, but the enterprise was a Java shop and programmability in
the native language was the key affordance.
## The Cost
The edge gateway we had built had hundreds of internal users and zero
external contributors. Best-in-class had hundreds of external
contributors solving problems before you knew you needed them solved.
"#;
let analysis = engine
.analyze(Ingest::text(text, Format::Markdown))
.next()
.expect("a stream of one")
.expect("analysis failed")
.analysis;
println!("Sentences: {}", analysis.total_sentences());
println!("Words: {}", analysis.total_words());
println!("Passive ratio: {:.1}%", analysis.passive_ratio() * 100.0);
println!(
"Vocabulary TTR: {:.2}",
analysis.vocabulary_ttr.unwrap_or(0.0)
);
println!(
"Nominalization: {:.1}%",
analysis.nominalization_ratio.unwrap_or(0.0) * 100.0
);
println!("\nSections:");
for section in &analysis.sections {
let heading = section.heading.as_deref().unwrap_or("(intro)");
let words: usize = section.paragraphs.iter().map(|p| p.word_count()).sum();
println!(
" {}: {} paragraphs, {} words",
heading,
section.paragraphs.len(),
words
);
}
}