Dataset, DataLoader & Klasifikasi Angka Tulisan Tangan

import torchvision.transforms as transforms # Pipeline preprocessing: jalankan berurutan transform = transforms.Compose([ transforms.ToTensor(), # PIL Image → Tensor [0, 1] transforms.Normalize( (0.1307,), # mean MNIST (0.3081,) # std MNIST ) ]) # Apa yang terjadi: # 1. ToTensor(): gambar 28×28 pixel (0-255) # → Tensor shape [1, 28, 28] dengan nilai [0.0, 1.0] # 2. Normalize(): (pixel - mean) / std # → Nilai terpusat di 0, spread ±1 # → Training lebih stabil & cepat converge

import torch from torch.utils.data import DataLoader import torchvision import torchvision.transforms as transforms # =========================== # 1. Transform pipeline # =========================== transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,)) ]) # =========================== # 2. Download & muat dataset # =========================== train_dataset = torchvision.datasets.MNIST( root='./data', train=True, # Training set (60K) download=True, # Download jika belum ada transform=transform ) test_dataset = torchvision.datasets.MNIST( root='./data', train=False, # Test set (10K) download=True, transform=transform ) print(f"Training: {len(train_dataset)} gambar") print(f"Test: {len(test_dataset)} gambar") # Training: 60000 gambar # Test: 10000 gambar # =========================== # 3. Buat DataLoader # =========================== train_loader = DataLoader( train_dataset, batch_size=64, # 64 gambar per batch shuffle=True # Acak setiap epoch ) test_loader = DataLoader( test_dataset, batch_size=64, shuffle=False # Test: tidak perlu acak ) # =========================== # 4. Intip satu batch # =========================== images, labels = next(iter(train_loader)) print(f"Batch images: {images.shape}") print(f"Batch labels: {labels.shape}") # Batch images: torch.Size([64, 1, 28, 28]) # ↑batch ↑channel ↑height ↑width # Batch labels: torch.Size([64]) # ↑64 angka (0-9)

import torch import torch.nn as nn class MNISTClassifier(nn.Module): def __init__(self): super().__init__() self.flatten = nn.Flatten() # [1,28,28] → [784] self.layers = nn.Sequential( nn.Linear(784, 256), # Hidden 1 nn.ReLU(), # Aktivasi nn.Linear(256, 128), # Hidden 2 nn.ReLU(), # Aktivasi nn.Linear(128, 10) # Output (10 kelas) ) def forward(self, x): x = self.flatten(x) # Flatten gambar x = self.layers(x) # Forward through layers return x model = MNISTClassifier() # Hitung total parameters total = sum(p.numel() for p in model.parameters()) print(f"Total parameters: {total:,}") # Total parameters: 235,146

import torch import torch.nn as nn from torch.utils.data import DataLoader import torchvision import torchvision.transforms as transforms # === Setup (dari section sebelumnya) === transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,)) ]) train_dataset = torchvision.datasets.MNIST('./data', train=True, download=True, transform=transform) test_dataset = torchvision.datasets.MNIST('./data', train=False, transform=transform) train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True) test_loader = DataLoader(test_dataset, batch_size=64) # === Model, Loss, Optimizer === model = MNISTClassifier() # Model dari section 5 loss_fn = nn.CrossEntropyLoss() # Untuk klasifikasi multi-kelas optimizer = torch.optim.Adam(model.parameters(), lr=0.001) # === Training Function === def train_one_epoch(model, loader, loss_fn, optimizer): model.train() # Mode training total_loss = 0 for images, labels in loader: pred = model(images) # Forward pass loss = loss_fn(pred, labels) # Hitung loss optimizer.zero_grad() # Reset gradient loss.backward() # Hitung gradient optimizer.step() # Update parameter total_loss += loss.item() return total_loss / len(loader) # === Evaluation Function === def evaluate(model, loader): model.eval() # Mode evaluasi correct = 0 total = 0 with torch.no_grad(): # Matikan gradient (hemat memory) for images, labels in loader: pred = model(images) _, predicted = pred.max(1) # Ambil kelas dengan skor tertinggi total += labels.size(0) correct += (predicted == labels).sum().item() return 100.0 * correct / total # === TRAINING! 🔥 === for epoch in range(10): loss = train_one_epoch(model, train_loader, loss_fn, optimizer) acc = evaluate(model, test_loader) print(f"Epoch {epoch+1:2d} | Loss: {loss:.4f} | Acc: {acc:.1f}%") # Output: # Epoch 1 | Loss: 0.3421 | Acc: 95.2% # Epoch 2 | Loss: 0.1543 | Acc: 96.4% # Epoch 3 | Loss: 0.1089 | Acc: 96.9% # Epoch 4 | Loss: 0.0842 | Acc: 97.2% # Epoch 5 | Loss: 0.0681 | Acc: 97.4% # Epoch 6 | Loss: 0.0554 | Acc: 97.5% # Epoch 7 | Loss: 0.0449 | Acc: 97.6% # Epoch 8 | Loss: 0.0378 | Acc: 97.7% # Epoch 9 | Loss: 0.0310 | Acc: 97.7% # Epoch 10 | Loss: 0.0254 | Acc: 97.8% 🎉

# =========================== # 1. Prediksi gambar individual # =========================== # Ambil satu gambar dari test set image, true_label = test_dataset[0] # Prediksi model.eval() with torch.no_grad(): output = model(image.unsqueeze(0)) # Tambah batch dimension probs = torch.nn.functional.softmax(output, dim=1) pred = output.argmax(dim=1).item() conf = probs[0][pred].item() * 100 print(f"True label: {true_label}") print(f"Predicted: {pred}") print(f"Confidence: {conf:.1f}%") # True label: 7 # Predicted: 7 # Confidence: 99.8% ✅ # =========================== # 2. Simpan model untuk nanti # =========================== torch.save(model.state_dict(), 'mnist_model.pth') print("Model tersimpan! 💾") # Muat kembali: loaded_model = MNISTClassifier() loaded_model.load_state_dict(torch.load('mnist_model.pth')) loaded_model.eval() print("Model dimuat kembali! ✅")