keyboard_arrow_down CNN

Examen Darwin Merchán Delgado

keyboard_arrow_down Red entrenada

Importando bibliotecas

import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import datasets, transforms

import numpy as np
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
import seaborn as sns

Inicializando parámetros para el uso de la GPU

use_cuda = torch.cuda.is_available()
device = torch.device("cuda" if use_cuda else "cpu")
train_kwargs = {'batch_size': 500}
test_kwargs = {'batch_size': 500}
if use_cuda:
cuda_kwargs = {'num_workers': 1, 'pin_memory': True, 'shuffle': True}
train_kwargs.update(cuda_kwargs)
test_kwargs.update(cuda_kwargs)

use_cuda

True

Descargando conjunto de datos (aplica transformaciones)

transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.5,), (0.3,)) ])

dataset1 = datasets.MNIST('../data', train=True, download=True, transform=transform)

dataset2 = datasets.MNIST('../data', train=False, transform=transform)
train_loader = torch.utils.data.DataLoader(dataset1, **train_kwargs)
test_loader = torch.utils.data.DataLoader(dataset2, **test_kwargs)

keyboard_arrow_down Red convolucional

Definición del modelo
 class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
self.dropout1 = nn.Dropout(0.25)
self.dropout2 = nn.Dropout(0.5)
self.fc1 = nn.Linear(9216, 128)
self.fc2 = nn.Linear(128, 10)

def forward(self, x):

x = F.relu(self.conv1(x))
x = F.relu(self.conv2(x))
x = F.max_pool2d(x, 2)
x = self.dropout1(x)
x = torch.flatten(x, 1)
x = F.relu(self.fc1(x))
x = self.dropout2(x)
x = self.fc2(x)
return F.log_softmax(x, dim=1)

def train(model, device, train_loader, optimizer, epoch):

model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if batch_idx % 20 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(epoch,
batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))

def test(model, device, test_loader):

model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += F.nll_loss(output, target, reduction='sum').item()
pred = output.argmax(dim=1, keepdim=True)
correct += pred.eq(target.view_as(pred)).sum().item()
test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\n'.format(test_loss,
correct, len(test_loader.dataset), 100. * correct / len(test_loader.dataset)))

Main execution

model = Net().to(device)
optimizer = torch.optim.Adadelta(model.parameters(), lr=1.0)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=1, gamma=0.7)
for epoch in range(5):
train(model, device, train_loader, optimizer, epoch)
test(model, device, test_loader)
scheduler.step()

keyboard_arrow_down Tarea 1
Entrenar la red convolucional pero esta vez con el conjunto de datos Fashion-MNIST

transform = transforms.Compose([ transforms.ToTensor(), transforms.Normalize((0.5,), (0.3,)) ])

dataset1 = datasets.FashionMNIST('../data', train=True, download=True, transform=transform)

dataset2 = datasets.FashionMNIST('../data', train=False, transform=transform)
train_loader = torch.utils.data.DataLoader(dataset1, **train_kwargs)
test_loader = torch.utils.data.DataLoader(dataset2, **test_kwargs)

Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz to ../data/FashionMNIST/raw/train-images-idx3-ubyte.gz
100%|██████████| 26421880/26421880 [00:01<00:00, 15864692.42it/s]
Extracting ../data/FashionMNIST/raw/train-images-idx3-ubyte.gz to ../data/FashionMNIST/raw

Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz to ../data/FashionMNIST/raw/train-labels-idx1-ubyte.gz
100%|██████████| 29515/29515 [00:00<00:00, 262657.37it/s]
 Extracting ../data/FashionMNIST/raw/train-labels-idx1-ubyte.gz to ../data/FashionMNIST/raw

Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz to ../data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz
100%|██████████| 4422102/4422102 [00:00<00:00, 4878946.00it/s]
Extracting ../data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz to ../data/FashionMNIST/raw

Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz
Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz to ../data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz
100%|██████████| 5148/5148 [00:00<00:00, 6258631.01it/s]Extracting ../data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz to ../data/FashionMNIST/raw

# TODO
class Net(nn.Module):
def __init__(self):
super(Net, self).__init__()
self.conv1 = nn.Conv2d(1, 32, 3, 1)
self.conv2 = nn.Conv2d(32, 64, 3, 1)
self.dropout1 = nn.Dropout(0.25)
self.dropout2 = nn.Dropout(0.5)
self.fc1 = nn.Linear(9216, 128)
self.fc2 = nn.Linear(128, 10)

def forward(self, x):

x = F.relu(self.conv1(x))
x = F.relu(self.conv2(x))
x = F.max_pool2d(x, 2)
x = self.dropout1(x)
x = torch.flatten(x, 1)
x = F.relu(self.fc1(x))
x = self.dropout2(x)
x = self.fc2(x)
return F.log_softmax(x, dim=1)

def train(model, device, train_loader, optimizer, epoch):

model.train()
for batch_idx, (data, target) in enumerate(train_loader):
data, target = data.to(device), target.to(device)
optimizer.zero_grad()
output = model(data)
loss = F.nll_loss(output, target)
loss.backward()
optimizer.step()
if batch_idx % 20 == 0:
print('Train Epoch: {} [{}/{} ({:.0f}%)]\tLoss: {:.6f}'.format(epoch,
batch_idx * len(data), len(train_loader.dataset),
100. * batch_idx / len(train_loader), loss.item()))

def test(model, device, test_loader):

model.eval()
test_loss = 0
correct = 0
with torch.no_grad():
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += F.nll_loss(output, target, reduction='sum').item()
pred = output.argmax(dim=1, keepdim=True)
correct += pred.eq(target.view_as(pred)).sum().item()
test_loss /= len(test_loader.dataset)
print('\nTest set: Average loss: {:.4f}, Accuracy: {}/{} ({:.2f}%)\n'.format(test_loss,
correct, len(test_loader.dataset), 100. * correct / len(test_loader.dataset)))

model = Net().to(device)
optimizer = torch.optim.Adadelta(model.parameters(), lr=1.0)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=1, gamma=0.7)
for epoch in range(50):
train(model, device, train_loader, optimizer, epoch)
test(model, device, test_loader)
scheduler.step()
 Train Epoch: 45 [0/60000 (0%)] Loss: 0.246246
Train Epoch: 45 [10000/60000 (17%)] Loss: 0.177380
Train Epoch: 45 [20000/60000 (33%)] Loss: 0.222282
Train Epoch: 45 [30000/60000 (50%)] Loss: 0.244674
Train Epoch: 45 [40000/60000 (67%)] Loss: 0.236790
Train Epoch: 45 [50000/60000 (83%)] Loss: 0.216829

Test set: Average loss: 0.2511, Accuracy: 9110/10000 (91.10%)

Train Epoch: 46 [0/60000 (0%)] Loss: 0.271067

Train Epoch: 46 [10000/60000 (17%)] Loss: 0.220825
Train Epoch: 46 [20000/60000 (33%)] Loss: 0.252584
Train Epoch: 46 [30000/60000 (50%)] Loss: 0.232903
Train Epoch: 46 [40000/60000 (67%)] Loss: 0.255602
Train Epoch: 46 [50000/60000 (83%)] Loss: 0.227497

Test set: Average loss: 0.2511, Accuracy: 9110/10000 (91.10%)

Train Epoch: 47 [0/60000 (0%)] Loss: 0.270406

Train Epoch: 47 [10000/60000 (17%)] Loss: 0.248372
Train Epoch: 47 [20000/60000 (33%)] Loss: 0.225195
Train Epoch: 47 [30000/60000 (50%)] Loss: 0.280196
Train Epoch: 47 [40000/60000 (67%)] Loss: 0.245476
Train Epoch: 47 [50000/60000 (83%)] Loss: 0.204588

Test set: Average loss: 0.2511, Accuracy: 9110/10000 (91.10%)

Train Epoch: 48 [0/60000 (0%)] Loss: 0.260848

Train Epoch: 48 [10000/60000 (17%)] Loss: 0.242858
Train Epoch: 48 [20000/60000 (33%)] Loss: 0.201093
Train Epoch: 48 [30000/60000 (50%)] Loss: 0.259808
Train Epoch: 48 [40000/60000 (67%)] Loss: 0.216152
Train Epoch: 48 [50000/60000 (83%)] Loss: 0.249747

Test set: Average loss: 0.2511, Accuracy: 9110/10000 (91.10%)

Train Epoch: 49 [0/60000 (0%)] Loss: 0.267562

Train Epoch: 49 [10000/60000 (17%)] Loss: 0.259132
Train Epoch: 49 [20000/60000 (33%)] Loss: 0.240020
Train Epoch: 49 [30000/60000 (50%)] Loss: 0.221690
Train Epoch: 49 [40000/60000 (67%)] Loss: 0.258446
Train Epoch: 49 [50000/60000 (83%)] Loss: 0.209825

Test set: Average loss: 0.2511, Accuracy: 9110/10000 (91.10%)

Reportar el desempeño alcanzado con la red convolucional en el conjunto de datos Fashion-MNIST

Entrenamiento con 5 epocas : Incia : 82.43% Termina : 90%

Entrenamiento con 20 epocas : Incia : 82.43% Termina : 90%

Entrenamiento con 50 epocas : Incia : 84.86% Termina : 91.10%, desde la epoca 24 marca 91,10 y no mejora el entrenamiento

Cargando conjunto de datos

y_pred = []
y_true = []

model.eval() # Poner el modelo en modo evaluación

with torch.no_grad(): # Desactivar el cálculo de gradientes
for data, target in test_loader:
data, target = data.to(device), target.to(device)
output = model(data)
pred = output.argmax(dim=1, keepdim=True) # Obtener el índice de la mayor probabilidad
y_pred.extend(pred.cpu().numpy())
y_true.extend(target.cpu().numpy())

Convertir listas a arrays de NumPy para compatibilidad con Scikit-Learn

y_pred = np.array(y_pred).flatten()
y_true = np.array(y_true).flatten()

Calcular la matriz de confusión

conf_mat = confusion_matrix(y_true, y_pred)

Graficar la matriz de confusión

plt.figure(figsize=(10, 7))
sns.heatmap(conf_mat, annot=True, fmt='d', cmap='Greens', xticklabels=dataset2.classes, yticklabels=dataset2.classes)
plt.xlabel('Prediccion')
plt.ylabel('Real')
plt.show()

Conclusión: En los entrenamiento podemos ver que mientras mas se entrene , mejor son los resultados como en Bag, no hay muchos errores.

Tarea 2
keyboard_arrow_down

[ ] ↳ 4 cells hidden