Replacing classifier head in PyTorch - Interactive Code Practice

Practice - 5 Tasks

1fill in blank

easy

Complete the code to replace the classifier head of a pretrained model with a new linear layer.

PyTorch

import torch.nn as nn
import torchvision
model = torchvision.models.resnet18(pretrained=True)
model.fc = nn.[1](512, 10)

Drag options to blanks, or click blank then click option'

AReLU

BConv2d

CBatchNorm2d

DLinear

Attempts:

3 left

💡 Hint

Common Mistakes

Using Conv2d instead of Linear for the classifier head.

Not matching the input feature size of the new layer.

The classifier head of ResNet18 is replaced by a linear layer with input features 512 and output features 10.

2fill in blank

medium

Complete the code to freeze all layers except the new classifier head.

PyTorch

for param in model.parameters():
    param.[1] = False
for param in model.fc.parameters():
    param.requires_grad = True

Drag options to blanks, or click blank then click option'

Arequires_grad

Bgrad

Crequires_grad_

Drequires_grad__

Attempts:

3 left

💡 Hint

Common Mistakes

Using incorrect attribute names like grad or requires_grad_.

Not freezing the pretrained layers before training.

Setting requires_grad to False freezes the parameters so they won't update during training.

3fill in blank

hard

Fix the error in the code to correctly replace the classifier head with a new linear layer.

PyTorch

import torchvision
import torch.nn as nn
model = torchvision.models.resnet50(pretrained=True)
model.fc = nn.Linear([1], 5)

Drag options to blanks, or click blank then click option'

A2048

B512

C1024

D256

Attempts:

3 left

💡 Hint

Common Mistakes

Using 512 which is the input size for ResNet18, not ResNet50.

Using arbitrary numbers without checking the model architecture.

ResNet50's classifier head input features size is 2048, so the new linear layer must match this.

4fill in blank

hard

Fill both blanks to create a new classifier head with dropout and linear layers.

PyTorch

model.fc = nn.Sequential(
    nn.Dropout(p=[1]),
    nn.Linear(512, [2])
)

Drag options to blanks, or click blank then click option'

A0.5

B10

C5

D0.3

Attempts:

3 left

💡 Hint

Common Mistakes

Using dropout probability greater than 1 or less than 0.

Mismatch between linear layer output size and number of classes.

Dropout with p=0.5 is common to reduce overfitting, and the linear layer outputs 10 classes.

5fill in blank

hard

Fill all three blanks to replace the classifier head and freeze pretrained layers except the new head.

PyTorch

import torchvision
import torch.nn as nn
model = torchvision.models.resnet34(pretrained=True)
for param in model.parameters():
    param.[1] = False
model.fc = nn.Linear([2], [3])
for param in model.fc.parameters():
    param.requires_grad = True

Drag options to blanks, or click blank then click option'

Arequires_grad

B512

C7

D1000

Attempts:

3 left

💡 Hint

Common Mistakes

Not freezing pretrained layers before training.

Using wrong input or output sizes for the linear layer.

Freezing pretrained layers requires setting requires_grad to False. ResNet34's fc input features are 512, and output is 7 classes.