View file src/colab/train_validation_test.py - Download

# -*- coding: utf-8 -*-
"""train_validation_test.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1FjJoQHYoL4K1cPtx4BXQ-pZbj0kw2l4s

Optimization of hyperparameter (learning rate)
"""

import torch
from torch import nn
from torch.utils.data import DataLoader
from torchvision import datasets, transforms
from torchvision.transforms import ToTensor
from torch.utils.data import TensorDataset

# Get cpu or gpu device for training.
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"Using {device} device")

torch.manual_seed(14)

transform = transforms.ToTensor()
training_data = datasets.MNIST(root='./data', train=True, download=True, transform=transform)
validation_and_test_data = datasets.MNIST(root='./data', train=False, download=True, transform=transform)

batch_size = 50

# Create data loaders.
train_dataloader = DataLoader(training_data, batch_size=batch_size)
validation_and_test_dataloader = DataLoader(validation_and_test_data, batch_size=batch_size)

train_dataset = train_dataloader.dataset

for X, y in validation_and_test_dataloader:
    print(f"Shape of X [N, C, H, W]: {X.shape}")
    print(f"Shape of y: {y.shape} {y.dtype}")
    break

print(f"train: {len(train_dataloader.dataset)}")
print(f"validation and test: {len(validation_and_test_dataloader.dataset)}")

validation_size = int(0.5 * len(validation_and_test_dataloader.dataset))
print(f"validation size: {validation_size}")
test_size = len(validation_and_test_dataloader.dataset) - validation_size
print(f"test size: {test_size}")
validation_dataset, test_dataset = torch.utils.data.random_split(validation_and_test_dataloader.dataset, [validation_size, test_size])

print(f"validation: {len(validation_dataset)}")
print(f"test: {len(test_dataset)}")

for batch, (X, y) in enumerate(train_dataloader):
    print(f"{batch}: {X.shape} {y.shape}")
    # X, y = X.to(device), y.to(device)
    break

for batch, (X, y) in enumerate(train_dataset):
    print(f"{batch}: {X.shape} {y}")
    break

validation_dataloader = DataLoader(validation_dataset, batch_size=batch_size, shuffle=True)
test_dataloader = DataLoader(test_dataset, batch_size=batch_size, shuffle=True)

for batch, (X, y) in enumerate(validation_dataloader):
    print(f"{batch}: {X.shape} {y.shape}")
    # X, y = X.to(device), y.to(device)
    break

# Define model
class NeuralNetwork(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear_relu_stack = nn.Sequential(
            nn.Linear(28*28, 512),
            nn.Sigmoid(),
            nn.Linear(512, 512),
            nn.Sigmoid(),
            nn.Linear(512, 10)
        )

    def set_lr(self, lr):
        self.optimizer = torch.optim.SGD(self.parameters(), lr=lr)

    def forward(self, x):
        logits = self.linear_relu_stack(x)
        return logits

    def zero_grad(self):
        self.optimizer.zero_grad()

    def optimize(self):
        self.optimizer.step()

model = NeuralNetwork().to(device)
print(model)

loss_fn = torch.nn.CrossEntropyLoss()

def train(dataloader, model, loss_fn, lr):
    size = len(dataloader.dataset)
    model.set_lr(lr)
    model.train()
    for batch, (X, y) in enumerate(dataloader):
        X, y = X.to(device), y.to(device)
        X = torch.flatten(X, 1, 3)  # transform X of shape 50, 1, 28, 28 into 50, 28*28
        # Compute prediction error
        pred = model(X)
        # Compute loss
        loss = loss_fn(pred, y)
        # Backpropagation
        model.zero_grad()
        loss.backward()
        # Optimize parameters
        model.optimize()
        if batch % 200 == 0:
            loss, current = loss.item(), batch * len(X)
            print(f"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]")

def test(dataloader, model, loss_fn, lr):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    model.set_lr(lr)
    model.eval()
    test_loss = 0
    correct = torch.tensor(0.0).requires_grad_(True)
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            X = torch.flatten(X, 1, 3)  # transform X of shape 50, 1, 28, 28 into 50, 28*28
            pred = model(X)
            test_loss += loss_fn(pred, y)
            correct += (pred.argmax(1) == y).type(torch.float32).sum().item()

    test_loss /= num_batches
    correct_rate = correct / size
    print(f"Test Error: \n Accuracy: {(100*correct_rate):>0.1f}%, Avg loss: {test_loss:>8f} \n")
    return correct_rate

epochs = 1
for t in range(epochs):
    print(f"Epoch {t+1}\n-------------------------------")
    train(train_dataloader, model, loss_fn, 0.1)
    test(validation_and_test_dataloader, model, loss_fn, 0.1)

print("Done!")

import math

def evaluate(llr, dataloader):
    lr = math.exp(llr)
    print(f"Evaluate for lr = {lr}")
    torch.manual_seed(14)
    model = NeuralNetwork().to(device)
    epochs = 1
    for t in range(epochs):
        print(f"Epoch {t+1}\n-------------------------------")
        train(train_dataloader, model, loss_fn, lr)
        correct = test(dataloader, model, loss_fn, lr)
    print(f"Evaluate for lr = {lr} : correct = {correct}")
    return correct

evaluate(0.1, validation_dataloader)

lr = torch.tensor(0.1).requires_grad_(True)
print(lr)
evaluate(lr, validation_dataloader)

llr = math.log(0.1)
eps = 0.01
g = 1

# while abs(g) > 0.001:
for i in range(5):
    print (f"*** STEP {i} ***")
    g = (evaluate(llr+eps, validation_dataloader) - evaluate(llr-eps, validation_dataloader)) / (2 * eps)  # the gradient
    llr = llr + 8 * g                                       # optimize lr
    print("")

lr = math.exp(llr)
print(f"Maximum correct for lr={lr} : validation : {evaluate(lr, validation_dataloader)}")
print(f"Maximum correct for lr={lr} : test       : {evaluate(lr, test_dataloader)}")