ai-0/lab4/src/main.rs

#![cfg_attr(not(debug_assertions), windows_subsystem = "windows")] // hide console window on Windows in release
#![allow(incomplete_features)]
#![feature(generic_const_exprs)]

mod algo;

use crate::algo::{gen_images, ComparationOperatorsModel};
use bgtu_ai_utility::gui::{boot_eframe, labeled_slider};
use eframe::egui;
use eframe::egui::Widget;
use egui_extras::{Column, TableBuilder};
use std::cmp::min;

fn main() -> eframe::Result {
    return boot_eframe(
        "Neural net",
        || MyApp::new()
    );
}


enum TrainingState {
    NoTrain,
    Training {
        epochs_left: usize,
        epochs_per_update: usize,
        n: f64,
    },
}

struct MyApp {
    model: ComparationOperatorsModel<7, 7, 8>,
    training: TrainingState,
    hidden_layer_size: usize,
    n: f64,
    epochs_count: usize,
    symbols: [&'static str; 8],
    compute_result: [f64; 8],
}

impl MyApp {
    fn new() -> Self {
        let imgs = gen_images();
        return Self {
            model: ComparationOperatorsModel::new(imgs.1),
            training: TrainingState::NoTrain,
            hidden_layer_size: 1,
            n: 0.1,
            epochs_count: 1,
            symbols: imgs.0,
            compute_result: [0.0; 8],
        };
    }
}


impl eframe::App for MyApp {
    fn update(&mut self, ui: &eframe::egui::Context, _frame: &mut eframe::Frame) {
        egui::CentralPanel::default().show(ui, |ui| {
            ui.add_enabled_ui(matches!(self.training, TrainingState::NoTrain), |ui| {
                labeled_slider(ui, "Hidden layer size", &mut self.hidden_layer_size, 1..=49, 1f64);
                if (self.hidden_layer_size != self.model.hidden_layer_size()) {
                    self.model.resize_hidden_layer(self.hidden_layer_size);
                }
                ui.label("");
                labeled_slider(ui, "η", &mut self.n, 0.0..=1.0, 0.001);
                ui.label("");
                labeled_slider(ui, "Epochs count", &mut self.epochs_count, 1..=500, 1f64);
                ui.label("");

                ui.horizontal(|ui| {
                    if ui.button("Train").clicked() {
                        self.training = TrainingState::Training {
                            epochs_left: self.epochs_count,
                            epochs_per_update: min(10, self.epochs_count / 60),
                            n: 0.0,
                        };
                        self.model.set_random_weights(&mut rand::rng());
                    }

                    match &mut self.training {
                        TrainingState::NoTrain => {
                            egui::ProgressBar::new(1.0).ui(ui);
                        }
                        TrainingState::Training {
                            epochs_left,
                            epochs_per_update,
                            n,
                        } => {
                            if *epochs_left >= *epochs_per_update {
                                for _ in 0..*epochs_per_update {
                                    self.model.train_epoch(*n);
                                }
                                *epochs_left -= *epochs_per_update;
                            } else {
                                for _ in 0..*epochs_left {
                                    self.model.train_epoch(*n);
                                }
                                *epochs_left = 0;
                            }

                            ui.add_enabled_ui(true, |ui| {
                                egui::ProgressBar::new(
                                    1.0 - (*epochs_left as f32) / (self.epochs_count as f32),
                                )
                                .ui(ui)
                            });

                            if 0 == *epochs_left {
                                self.training = TrainingState::NoTrain;
                            }
                        }
                    }
                })
            });

            ui.label("");

            ui.horizontal(|ui| {
                for (i, s) in self.symbols.iter().enumerate() {
                    if ui.button(s.to_owned()).clicked() {
                        self.model.set_input_from_etalon(i);
                    }
                }
            });

            ui.label("");

            for y in 0..7 {
                ui.horizontal(|ui| {
                    for x in 0..7 {
                        ui.checkbox(&mut self.model.input[y][x], "");
                    }
                });
            }

            ui.label("");
            ui.add_enabled_ui(matches!(self.training, TrainingState::NoTrain), |ui| {
                if ui.button("Check").clicked() {
                    self.compute_result = self.model.calculate_predictions();
                }
            });

            TableBuilder::new(ui)
                .striped(true) // Alternating row colors
                .resizable(true)
                .column(Column::remainder())
                .column(Column::remainder())
                .header(20.0, |mut header| {
                    header.col(|ui| {
                        ui.label("Class");
                    });
                    header.col(|ui| {
                        ui.label("Probability");
                    });
                })
                .body(|body| {
                    body.rows(20.0, self.symbols.len(), |mut row| {
                        let i = row.index();
                        row.col(|ui| {
                            ui.label(self.symbols[i]);
                        });
                        row.col(|ui| {
                            ui.label(self.compute_result[i].to_string());
                        });
                    });
                });
        });

        if matches!(self.training, TrainingState::Training {..}) {
            ui.request_repaint()
        }
    }
}