Train

In [ ]:

import time
import torch
def loss_f(y_pred, y_true):
    return torch.nn.functional.cross_entropy(y_pred, y_true)

import time import torch def loss_f(y_pred, y_true): return torch.nn.functional.cross_entropy(y_pred, y_true)

In [ ]:

def train(model, optimizer, train_data, valid_data, device, epochs, torch_resize):
    model.to(device)
    loss_history = []
    acc_history = []
    
    startTime = time.time()
    for e in range(epochs):
        model.train()
        for x, y in train_data:
            x = torch_resize(x).to(device)
            y = y.to(device)
            y_pred = model.forward(x)

            loss = loss_f(y_pred, y)
            optimizer.zero_grad()
            loss.backward()
            optimizer.step()
        
        loss_sum, acc = get_acc(model, valid_data, device, torch_resize)
        loss_history.append(loss_sum)
        acc_history.append(acc)
        print(f"Epoch{e+1} loss {loss_sum} valid accuracy {acc}%")
            
    endTime = time.time()
    print(f'Train cost {(endTime - startTime)}s')
    return loss_history,acc_history

def train(model, optimizer, train_data, valid_data, device, epochs, torch_resize): model.to(device) loss_history = [] acc_history = [] startTime = time.time() for e in range(epochs): model.train() for x, y in train_data: x = torch_resize(x).to(device) y = y.to(device) y_pred = model.forward(x) loss = loss_f(y_pred, y) optimizer.zero_grad() loss.backward() optimizer.step() loss_sum, acc = get_acc(model, valid_data, device, torch_resize) loss_history.append(loss_sum) acc_history.append(acc) print(f"Epoch{e+1} loss {loss_sum} valid accuracy {acc}%") endTime = time.time() print(f'Train cost {(endTime - startTime)}s') return loss_history,acc_history

In [ ]:

def get_acc(model, data, device, torch_resize):
    model.eval()
    loss_sum = 0
    num_true, num_total = 0, 0
    with torch.no_grad():
        for x, y in data:
            x = torch_resize(x).to(device)
            y = y.to(device)
            y_pred = model.forward(x)

            loss = loss_f(y_pred, y)
            loss_sum += loss.to(device)
            num_total += y.size(0)
            num_true += (torch.argmax(y_pred, dim=1) == y).sum().to(device)
    
    return loss_sum, num_true / num_total * 100

def get_acc(model, data, device, torch_resize): model.eval() loss_sum = 0 num_true, num_total = 0, 0 with torch.no_grad(): for x, y in data: x = torch_resize(x).to(device) y = y.to(device) y_pred = model.forward(x) loss = loss_f(y_pred, y) loss_sum += loss.to(device) num_total += y.size(0) num_true += (torch.argmax(y_pred, dim=1) == y).sum().to(device) return loss_sum, num_true / num_total * 100