src/train_efficientnet.py

import random
import numpy as np
import torch
import torch.nn as nn
import torch.optim as optim
import matplotlib.pyplot as plt

from torchvision.models import efficientnet_b0, EfficientNet_B0_Weights
from tqdm import tqdm

from src.dataset import create_dataloaders


def set_seed(seed=42):
    random.seed(seed)
    np.random.seed(seed)
    torch.manual_seed(seed)
    torch.cuda.manual_seed_all(seed)

#training loop
def train_one_epoch(model, dataloader, criterion, optimizer, device):
    model.train()
    total_loss = 0.0
    correct = 0
    total = 0

    for images, labels, _ in tqdm(dataloader):
        images = images.to(device)
        labels = labels.to(device).long()

        optimizer.zero_grad()
        outputs = model(images)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        total_loss += loss.item()
        preds = torch.argmax(outputs, dim=1)
        correct += (preds == labels).sum().item()
        total += labels.size(0)

    avg_loss = total_loss / len(dataloader)
    accuracy = 100.0 * correct / total
    return avg_loss, accuracy

#validation loop
def validate(model, dataloader, criterion, device):
    model.eval()
    total_loss = 0.0
    correct = 0
    total = 0

    with torch.no_grad():
        for images, labels, _ in dataloader:
            images = images.to(device)
            labels = labels.to(device).long()

            outputs = model(images)
            loss = criterion(outputs, labels)

            total_loss += loss.item()
            preds = torch.argmax(outputs, dim=1)
            correct += (preds == labels).sum().item()
            total += labels.size(0)

    avg_loss = total_loss / len(dataloader)
    accuracy = 100.0 * correct / total
    return avg_loss, accuracy


def plot_curves(train_losses, val_losses, train_accuracies, val_accuracies):
    epochs_done = len(train_losses)

    plt.figure(figsize=(12, 5))

    plt.subplot(1, 2, 1)
    plt.plot(range(1, epochs_done + 1), train_losses, marker="o", label="Train Loss")
    plt.plot(range(1, epochs_done + 1), val_losses, marker="o", label="Val Loss")
    plt.xlabel("Epoch")
    plt.ylabel("Loss")
    plt.title("Loss Curve")
    plt.legend()

    plt.subplot(1, 2, 2)
    plt.plot(range(1, epochs_done + 1), train_accuracies, marker="o", label="Train Accuracy")
    plt.plot(range(1, epochs_done + 1), val_accuracies, marker="o", label="Val Accuracy")
    plt.xlabel("Epoch")
    plt.ylabel("Accuracy (%)")
    plt.title("Accuracy Curve")
    plt.legend()

    plt.show()


set_seed(42)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)

train_loader, val_loader, test_loader, class_names = create_dataloaders()
num_classes = len(class_names)

print("Number of classes:", num_classes)
print("Classes:", class_names)

weights = EfficientNet_B0_Weights.DEFAULT
model = efficientnet_b0(weights=weights)

#freezing the feature extractor and modifying the classifier
for param in model.features.parameters():
    param.requires_grad = False

in_features = model.classifier[1].in_features
model.classifier = nn.Sequential(
    nn.Dropout(p=0.3),
    nn.Linear(in_features, num_classes)
)

model = model.to(device)

criterion = nn.CrossEntropyLoss()

train_losses = []
val_losses = []
train_accuracies = []
val_accuracies = []

# Phase 1: train classifier only
optimizer = optim.AdamW(
    filter(lambda p: p.requires_grad, model.parameters()),
    lr=1e-3,
    weight_decay=1e-4
)

epochs_phase1 = 10

for epoch in range(epochs_phase1):
    train_loss, train_acc = train_one_epoch(model, train_loader, criterion, optimizer, device)
    val_loss, val_acc = validate(model, val_loader, criterion, device)

    train_losses.append(train_loss)
    val_losses.append(val_loss)
    train_accuracies.append(train_acc)
    val_accuracies.append(val_acc)

    print(
        f"[Phase 1] Epoch {epoch + 1}/{epochs_phase1} | "
        f"Train Loss: {train_loss:.4f}, Train Acc: {train_acc:.2f}% | "
        f"Val Loss: {val_loss:.4f}, Val Acc: {val_acc:.2f}%"
    )

# Unfreeze all feature layers for full fine-tuning
for param in model.features.parameters():
    param.requires_grad = True

optimizer = optim.AdamW(
    model.parameters(),
    lr=1e-5,
    weight_decay=1e-4
)

epochs_phase2 = 20

for epoch in range(epochs_phase2):
    train_loss, train_acc = train_one_epoch(model, train_loader, criterion, optimizer, device)
    val_loss, val_acc = validate(model, val_loader, criterion, device)

    train_losses.append(train_loss)
    val_losses.append(val_loss)
    train_accuracies.append(train_acc)
    val_accuracies.append(val_acc)

    print(
        f"[Phase 2] Epoch {epoch + 1}/{epochs_phase2} | "
        f"Train Loss: {train_loss:.4f}, Train Acc: {train_acc:.2f}% | "
        f"Val Loss: {val_loss:.4f}, Val Acc: {val_acc:.2f}%"
    )

plot_curves(train_losses, val_losses, train_accuracies, val_accuracies)