Alignment-Lab-AI's picture
Upload folder using huggingface_hub
d5bfab8 verified
use super::{AnalyticsDirectory, BatchProgramAnalyzerPlugin, BatchProgramAnalyzerContext};
use crate::common::create_csv_file;
use crate::common::RecordBigram;
use crate::common::RecordTrigram;
use crate::common::RecordSkipgram;
use crate::common::RecordUnigram;
use loda_rust_core;
use loda_rust_core::parser::{InstructionId, InstructionParameter, ParsedProgram};
use std::path::PathBuf;
use std::error::Error;
use std::collections::HashMap;
type HistogramBigramKey = (String,String);
type HistogramTrigramKey = (String,String,String);
type HistogramSkipgramKey = (String,String);
/// Creates [N-gram] csv files with LODA source parameters.
///
/// Traverses all the programs inside the `loda-programs/oeis` dir.
/// It looks for all the LODA assembly programs there are.
/// Determines the most frequent combinations of source parameters.
///
/// ---
///
/// This outputs a `histogram_source_unigram.csv` file, with this format:
///
/// ```csv
/// count;word
/// 188414;1
/// 144834;NONE
/// 125947;$0
/// 116255;$1
/// 107093;2
/// 101601;STOP
/// 101601;START
/// 87819;$2
/// 53412;$3
/// 39696;$4
/// ```
///
/// Learnings from this unigram with LODA programs:
///
/// Learning A: Source is most often a contant.
///
/// Learning B: Often there is no source parameter, such as for loop begin `lpb $3` or loop end `lpe`.
///
/// Learning C: The source register `$0` is most used.
///
/// ---
///
/// This outputs a `histogram_source_bigram.csv` file, with this format:
///
/// ```csv
/// count;word0;word1
/// 45608;NONE;1
/// 36119;$1;STOP
/// 35448;1;NONE
/// 31346;START;1
/// 28262;NONE;$1
/// 25464;START;$0
/// 20347;1;$0
/// 20234;1;1
/// 18594;$0;2
/// ```
///
/// Learnings from this bigram with LODA programs:
///
/// Learning A: Source constant is most likely to be followed by another source constant.
///
/// Learning B: Source register `$0` is most likely to be followed by a source constant.
///
/// Learning C: The program is most likely to start with a source constant.
///
/// Learning D: The program is most likely to stop with a source constant.
///
/// ---
///
/// This outputs a `histogram_source_trigram.csv` file, with this format:
///
/// ```csv
/// count;word0;word1;word2
/// 15775;1;NONE;1
/// 13470;NONE;$1;STOP
/// 9546;NONE;1;$1
/// 8145;START;1;1
/// 7357;START;1;NONE
/// 7331;NONE;1;$2
/// 7266;1;1;NONE
/// 6969;START;$0;2
/// 6467;START;1;$0
/// ```
///
/// Learnings from this trigram with LODA programs:
///
/// Learning A: Three source contants in succession are most popular.
///
/// Learning B: Programs often start with two source contants.
///
/// Learning C: The source register `$0` is often surrounded by two source constants.
///
/// ---
///
/// This outputs a `histogram_source_skipgram.csv` file, with this format:
///
/// ```csv
/// count;word0;word2
/// 30765;1;NONE
/// 28622;1;1
/// 28617;NONE;STOP
/// 24477;START;1
/// 20608;START;$0
/// 19542;$0;1
/// 18662;1;$0
/// 18579;$0;NONE
/// ```
///
/// Learnings from this skipgram with LODA programs:
///
/// Learning A: Constant and some junk is usually followed by another constant.
///
/// Learning B: Program start and some junk is usually followed a constant.
///
/// Learning C: The `$0` and some junk is usually followed by a constant.
///
/// [N-gram]: <https://en.wikipedia.org/wiki/N-gram>
pub struct AnalyzeSourceNgram {
analytics_directory: AnalyticsDirectory,
histogram_unigram: HashMap<String,u32>,
histogram_bigram: HashMap<HistogramBigramKey,u32>,
histogram_trigram: HashMap<HistogramTrigramKey,u32>,
histogram_skipgram: HashMap<HistogramSkipgramKey,u32>,
}
impl AnalyzeSourceNgram {
pub fn new(analytics_directory: AnalyticsDirectory) -> Self {
Self {
analytics_directory,
histogram_unigram: HashMap::new(),
histogram_bigram: HashMap::new(),
histogram_trigram: HashMap::new(),
histogram_skipgram: HashMap::new(),
}
}
fn extract_words(parsed_program: &ParsedProgram) -> Vec<String> {
let mut words: Vec<String> = vec!();
words.push("START".to_string());
for instruction in &parsed_program.instruction_vec {
if instruction.parameter_vec.len() < 2 {
words.push("NONE".to_string());
continue;
}
match instruction.instruction_id {
InstructionId::EvalSequence |
InstructionId::LoopBegin |
InstructionId::LoopEnd |
InstructionId::UnofficialFunction { .. } |
InstructionId::UnofficialLoopBeginSubtract => {
words.push("NONE".to_string());
continue;
},
_ => {}
}
let parameter: &InstructionParameter = &instruction.parameter_vec[1];
words.push(parameter.to_string());
}
words.push("STOP".to_string());
words
}
fn populate_unigram(&mut self, words: &Vec<String>) {
let keys: Vec<String> = words.clone();
for key in keys {
let counter = self.histogram_unigram.entry(key).or_insert(0);
*counter += 1;
}
}
fn populate_bigram(&mut self, words: &Vec<String>) {
let mut keys = Vec::<HistogramBigramKey>::new();
let mut prev_word = String::new();
for (index, word1) in words.iter().enumerate() {
let word0: String = prev_word;
prev_word = word1.clone();
if index == 0 {
continue;
}
let key: HistogramBigramKey = (word0, word1.clone());
keys.push(key);
}
for key in keys {
let counter = self.histogram_bigram.entry(key).or_insert(0);
*counter += 1;
}
}
fn populate_trigram(&mut self, words: &Vec<String>) {
let mut keys = Vec::<HistogramTrigramKey>::new();
let mut prev_prev_word = String::new();
let mut prev_word = String::new();
for (index, word2) in words.iter().enumerate() {
let word0: String = prev_prev_word;
let word1: String = prev_word.clone();
prev_prev_word = prev_word;
prev_word = word2.clone();
if index < 2 {
continue;
}
let key: HistogramTrigramKey = (word0, word1, word2.clone());
keys.push(key);
}
for key in keys {
let counter = self.histogram_trigram.entry(key).or_insert(0);
*counter += 1;
}
}
fn populate_skipgram(&mut self, words: &Vec<String>) {
let mut keys = Vec::<HistogramSkipgramKey>::new();
let mut prev_prev_word = String::new();
let mut prev_word = String::new();
for (index, word2) in words.iter().enumerate() {
let word0: String = prev_prev_word;
prev_prev_word = prev_word;
prev_word = word2.clone();
if index < 2 {
continue;
}
let key: HistogramSkipgramKey = (word0, word2.clone());
keys.push(key);
}
for key in keys {
let counter = self.histogram_skipgram.entry(key).or_insert(0);
*counter += 1;
}
}
fn save_unigram(&self) -> Result<(), Box<dyn Error>> {
// Convert from dictionary to array
let mut records = Vec::<RecordUnigram>::new();
for (histogram_key, histogram_count) in &self.histogram_unigram {
let record = RecordUnigram {
count: *histogram_count,
word: histogram_key.clone(),
};
records.push(record);
}
// Move the most frequently occuring items to the top
// Move the lesser used items to the bottom
records.sort_unstable_by_key(|item| (item.count, item.word.clone()));
records.reverse();
// Save as a CSV file
let output_path: PathBuf = self.analytics_directory.histogram_source_unigram_file();
create_csv_file(&records, &output_path)
}
fn save_bigram(&self) -> Result<(), Box<dyn Error>> {
// Convert from dictionary to array
let mut records = Vec::<RecordBigram>::new();
for (histogram_key, histogram_count) in &self.histogram_bigram {
let record = RecordBigram {
count: *histogram_count,
word0: histogram_key.0.clone(),
word1: histogram_key.1.clone()
};
records.push(record);
}
// Move the most frequently occuring items to the top
// Move the lesser used items to the bottom
records.sort_unstable_by_key(|item| (item.count, item.word0.clone(), item.word1.clone()));
records.reverse();
// Save as a CSV file
let output_path: PathBuf = self.analytics_directory.histogram_source_bigram_file();
create_csv_file(&records, &output_path)
}
fn save_trigram(&self) -> Result<(), Box<dyn Error>> {
// Convert from dictionary to array
let mut records = Vec::<RecordTrigram>::new();
for (histogram_key, histogram_count) in &self.histogram_trigram {
let record = RecordTrigram {
count: *histogram_count,
word0: histogram_key.0.clone(),
word1: histogram_key.1.clone(),
word2: histogram_key.2.clone()
};
records.push(record);
}
// Move the most frequently occuring items to the top
// Move the lesser used items to the bottom
records.sort_unstable_by_key(|item| (item.count, item.word0.clone(), item.word1.clone(), item.word2.clone()));
records.reverse();
// Save as a CSV file
let output_path: PathBuf = self.analytics_directory.histogram_source_trigram_file();
create_csv_file(&records, &output_path)
}
fn save_skipgram(&self) -> Result<(), Box<dyn Error>> {
// Convert from dictionary to array
let mut records = Vec::<RecordSkipgram>::new();
for (histogram_key, histogram_count) in &self.histogram_skipgram {
let record = RecordSkipgram {
count: *histogram_count,
word0: histogram_key.0.clone(),
word2: histogram_key.1.clone()
};
records.push(record);
}
// Move the most frequently occuring items to the top
// Move the lesser used items to the bottom
records.sort_unstable_by_key(|item| (item.count, item.word0.clone(), item.word2.clone()));
records.reverse();
// Save as a CSV file
let output_path: PathBuf = self.analytics_directory.histogram_source_skipgram_file();
create_csv_file(&records, &output_path)
}
}
impl BatchProgramAnalyzerPlugin for AnalyzeSourceNgram {
fn plugin_name(&self) -> &'static str {
"AnalyzeSourceNgram"
}
fn analyze(&mut self, context: &BatchProgramAnalyzerContext) -> Result<(), Box<dyn Error>> {
let words: Vec<String> = Self::extract_words(&context.parsed_program);
self.populate_unigram(&words);
self.populate_bigram(&words);
self.populate_trigram(&words);
self.populate_skipgram(&words);
Ok(())
}
fn save(&self) -> Result<(), Box<dyn Error>> {
self.save_unigram()?;
self.save_bigram()?;
self.save_trigram()?;
self.save_skipgram()?;
Ok(())
}
fn human_readable_summary(&self) -> String {
let rows: Vec<String> = vec![
format!("unigram: {:?}", self.histogram_unigram.len()),
format!("bigram: {:?}", self.histogram_bigram.len()),
format!("trigram: {:?}", self.histogram_trigram.len()),
format!("skipgram: {:?}", self.histogram_skipgram.len()),
];
rows.join(", ")
}
}