Model playground

This notebook is for testing out different models and architectures. We have loaded a dataset for you and defined a DataLoader for it. The goal is to train a model that achieves at least 75% accuracy on the test set.

You can use any model architecture, any optimizer, and any loss function.

We expect you to try out: - Regularization techniques: dropout, weight decay, early stopping, data augmentation - Transfer learning - Hyperparameter tuning

We have provided you with some code to get started. Feel free to modify it.

import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torch.utils.data import sampler

import torchvision.datasets as dset
import torchvision.transforms as T
import matplotlib.pyplot as plt

import numpy as np

USE_GPU = True
dtype = torch.float32 # We will be using float throughout this tutorial.

if USE_GPU and torch.cuda.is_available():
    device = torch.device('cuda')
else:
    device = torch.device('cpu')

# Constant to control how frequently we print train loss.
print_every = 100
print('using device:', device)

using device: cuda

from medmnist import DermaMNIST
transform = T.Compose([
                T.ToTensor(),
                T.Normalize((0.5719),(0.1684))
            ])

dataset_train = DermaMNIST(root='utils/datasets', split='train', transform=transform, download=True)
loader_train = DataLoader(dataset_train, batch_size=64, shuffle=True, num_workers=2)

dataset_val = DermaMNIST(root='utils/datasets', split='val', transform=transform, download=True)
loader_val = DataLoader(dataset_val, batch_size=64, shuffle=True, num_workers=2)

dataset_test = DermaMNIST(root='utils/datasets', split='test', transform=transform, download=True)
loader_test = DataLoader(dataset_test, batch_size=64, shuffle=True, num_workers=2)

Downloading https://zenodo.org/records/10519652/files/dermamnist.npz?download=1 to cm2003/datasets/dermamnist.npz

100%|██████████| 19725078/19725078 [00:00<00:00, 32471604.97it/s]

Using downloaded and verified file: cm2003/datasets/dermamnist.npz
Using downloaded and verified file: cm2003/datasets/dermamnist.npz

image, label = dataset_train[5]
print(f"Image shape: {image.shape}")

# Reshape and convert the image for display
image_display = image.permute(1, 2, 0).numpy()  # Change from (3, 28, 28) to (28, 28, 3)
image_display = (image_display * 0.1684) + 0.5719  # Denormalize the image
image_display = np.clip(image_display, 0, 1)  # Clip values to [0, 1] range

plt.imshow(image_display)
plt.title(f'Label: {label}')
plt.axis('off')
plt.show()

Image shape: torch.Size([3, 28, 28])

from utils.models import alexnet
from utils.train import train, check_accuracy


model = alexnet.AlexNet(num_classes=7, in_channels=3)

# Define loss function and optimizer
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=0.001)

# Set device
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Set number of epochs
num_epochs = 10


# Train the model
model, results = train(model, loader_train, loader_val, criterion, optimizer, device, num_epochs)

# Check accuracy on the test set
test_loss, test_accuracy = check_accuracy(model, loader_test, criterion, device)
train_loss, train_accuracy = check_accuracy(model, loader_train, criterion, device)
print(f"Test Loss: {test_loss:.4f}, Test Accuracy: {test_accuracy:.4f}")
print(f"Train Loss: {train_loss:.4f}, Train Accuracy: {train_accuracy:.4f}")

Epoch 1/10:
Train Loss: 1.0551
Val Loss: 0.9193, Val Accuracy: 0.6680
Epoch 2/10:
Train Loss: 0.9071
Val Loss: 0.8139, Val Accuracy: 0.6909
Epoch 3/10:
Train Loss: 0.8440
Val Loss: 0.7856, Val Accuracy: 0.7069
Epoch 4/10:
Train Loss: 0.8109
Val Loss: 0.7826, Val Accuracy: 0.7049
Epoch 5/10:
Train Loss: 0.7950
Val Loss: 0.7436, Val Accuracy: 0.7188
Epoch 6/10:
Train Loss: 0.7606
Val Loss: 0.7449, Val Accuracy: 0.7228
Epoch 7/10:
Train Loss: 0.7473
Val Loss: 0.7118, Val Accuracy: 0.7458
Epoch 8/10:
Train Loss: 0.7360
Val Loss: 0.6961, Val Accuracy: 0.7468
Epoch 9/10:
Train Loss: 0.7147
Val Loss: 0.7115, Val Accuracy: 0.7468
Epoch 10/10:
Train Loss: 0.7004
Val Loss: 0.7092, Val Accuracy: 0.7318
Test Loss: 0.7138, Test Accuracy: 0.7232
Train Loss: 0.6663, Train Accuracy: 0.7421