# PyTorch Assertions

In this notebook, you'll learn how to use most used PyTorch assertion functions.

assert_pytorch_tensor_variable_name_equals_serialized(student_variable_name, location): Checks if student tensor object is equals to expected tensor. The expected tensor is stored in a pickle file.
assert_pytorch_tensor_variable_name_is_in_expected_device(student_variable_name, expected_device): Checks if student tensor object is in expected device. The expected device is a string.
assert_vision_transform_variable_name_equals_serialized(student_variable_name, location, **kwargs): Checks if student transformation pipeline object is equals to expected transformation pipeline. The expected transformation pipeline is stored in a pickle file.
assert_dataset_vision_image_folder_variable_name_equals_serialized(student_variable_name, location): Checks if student dataset(image directory) image folder object is equals to expected dataset image folder. The expected dataset image folder is stored in a pickle file.

Load the utils.py file to use the assertion functions.

exec(open("utils.py").read())

# Activities

Now, with activities examples, you'll learn how to use the assertion functions.

# 1: Create a 2d PyTorch tensor.

Use below list to create a 2d PyTorch tensor ans store it in a variable student.

data = [[1, 2, 3], [4, 5, 6]]

import torch
student = ...

Solution:

import torch
data = [[1, 2, 3], [4, 5, 6]]
student = torch.tensor(data)
print(student)

As we have created a tensor, save the tensor in a file using below code in a pickle file.

import pickle
with open('activity_solutions_files/expected_tensor_1.pt', 'wb') as f:
    pickle.dump(student, f)

or, serialize the tensor using the below code.

location = 'expected_tensor_1.pt' # Using this, we don't need to mention the activity_solutions_files path in the assertion.
serialize_pickle(student, location)

assert_pytorch_tensor_variable_name_equals_serialized('student', 'expected_tensor_1.pt')

# 2: Change the device of the tensor.

Change the device of the tensor to cuda and store it in a variable student.

student = student...

Solution:

student = student.to('cuda')

As we have changed the device of the tensor, save the tensor in a file using below code in a pickle file.

import pickle
with open('activity_solutions_files/expected_tensor_2.pt', 'wb') as f:
    pickle.dump(student, f)

Assertions:

assert_pytorch_tensor_equals_serialized(student, 'expected_tensor_2.pt')

# 3: Test the transformation pipeline.

If student create a transformation pipeline with below parameters, then assert it with the expected transformation pipeline.

import torchvision.transforms as transforms
student = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

Solution:

import torchvision.transforms as transforms
student = transforms.Compose([
    transforms.RandomResizedCrop(224),
    transforms.RandomHorizontalFlip(),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
])

As we have created a transformation pipeline, save the transformation pipeline in a file using below code in a pickle file.

location = 'expected_transform_1.pt' # Using this, we don't need to mention the activity_solutions_files path in the assertion.
serialize_pickle(student, location)

Assertions:

assert_vision_transform_variable_name_equals_serialized('student', 'expected_transform_1.pt')

# 4: Test the dataset image folder.

If student create a dataset image folder with below parameters, then assert it with the expected dataset image folder.

import torchvision.datasets as datasets
student = datasets.ImageFolder(root='path/to/data', transform=student)

Here, the path/to/data is the path to the image folder and student is the transformation pipeline created in the previous activity example.

Solution:

import torchvision.datasets as datasets
student = datasets.ImageFolder(root='path/to/data', transform=student)

As we have created a dataset image folder, save the dataset image folder in a file using below code in a pickle file.

location = 'expected_dataset_image_folder_1.pt' # Using this, we don't need to mention the activity_solutions_files path in the assertion.
serialize_pickle(student, location)

Assertions:

assert_dataset_vision_image_folder_variable_name_equals_serialized('student', 'expected_dataset_image_folder_1.pt')