team-2/test2_with_training/scripts/train.py

#!/usr/bin/env python3
# scripts/train.py
"""
Script di training per il classificatore basilico vs pomodoro.
Struttura:
  - carica dataset da data/basil_tomato/train e /val
  - transfer learning con EfficientNet-B0
  - salva il miglior modello in models/basil_tomato_classifier.pth
"""
import os
import torch
import torch.nn as nn
import torch.optim as optim
from torchvision import datasets, transforms, models
from torch.utils.data import DataLoader

# 1) Percorsi dataset
train_dir = "data/basil_tomato/train"
val_dir   = "data/basil_tomato/val"

# 2) Trasformazioni dati
train_transforms = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
val_transforms = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])

# 3) Crea dataset e DataLoader
train_ds = datasets.ImageFolder(train_dir, transform=train_transforms)
val_ds   = datasets.ImageFolder(val_dir,   transform=val_transforms)

train_loader = DataLoader(train_ds, batch_size=32, shuffle=True,  num_workers=4)
val_loader   = DataLoader(val_ds,   batch_size=32, shuffle=False, num_workers=4)

print(f"Classi trovate: {train_ds.classes}")
print(f"Numero immagini train: {len(train_ds)}, validation: {len(val_ds)}")

# 4) Configura device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")

# 5) Costruisci il modello
model = models.efficientnet_b0(pretrained=True)
num_classes = len(train_ds.classes)
model.classifier[1] = nn.Linear(model.classifier[1].in_features, num_classes)
model = model.to(device)

# 6) Definisci criterio e ottimizzatore
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=1e-4, weight_decay=1e-5)

# 7) Funzioni di training e validation

def train_epoch():
    model.train()
    running_loss, running_corrects = 0.0, 0

    for inputs, labels in train_loader:
        inputs, labels = inputs.to(device), labels.to(device)
        optimizer.zero_grad()
        outputs = model(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        running_loss     += loss.item() * inputs.size(0)
        running_corrects += (outputs.argmax(1) == labels).sum().item()

    epoch_loss = running_loss / len(train_ds)
    epoch_acc  = running_corrects / len(train_ds)
    return epoch_loss, epoch_acc


def validate_epoch():
    model.eval()
    val_loss, val_corrects = 0.0, 0

    with torch.no_grad():
        for inputs, labels in val_loader:
            inputs, labels = inputs.to(device), labels.to(device)
            outputs = model(inputs)
            loss = criterion(outputs, labels)

            val_loss     += loss.item() * inputs.size(0)
            val_corrects += (outputs.argmax(1) == labels).sum().item()

    loss = val_loss / len(val_ds)
    acc  = val_corrects / len(val_ds)
    return loss, acc

# 8) Training loop principale
best_val_acc = 0.0
os.makedirs("models", exist_ok=True)

for epoch in range(1, 11):  # 10 epoche
    train_loss, train_acc = train_epoch()
    val_loss,   val_acc   = validate_epoch()

    print(f"Epoca {epoch}: train_loss={train_loss:.4f}, train_acc={train_acc:.4f} | "
          f"val_loss={val_loss:.4f}, val_acc={val_acc:.4f}")

    # Salva il modello migliore
    if val_acc > best_val_acc:
        best_val_acc = val_acc
        save_path = os.path.join("models", "basil_tomato_classifier.pth")
        torch.save(model.state_dict(), save_path)
        print(f"--> Nuovo best model salvato con val_acc={val_acc:.4f}")

print("Training completato. Best val_acc: {:.4f}".format(best_val_acc))