Neural Network Programming - Deep Learning with PyTorch

with deeplizard.

PyTorch Datasets and DataLoaders - Training Set Exploration for Deep Learning and AI

October 31, 2018 by

Blog

PyTorch Datasets and DataLoaders for deep Learning

Welcome back to this series on neural network programming with PyTorch. In this post, we see how to work with the Dataset and DataLoader PyTorch classes.

ai cyborg

Our goal in this post is to get comfortable using the dataset and data loader objects as well as to get a feel for our training set. Without further ado, let’s get started.

From a high-level perspective, we are still in the preparing data stage of our deep learning project.

  • Prepare the data
  • Build the model
  • Train the model
  • Analyze the model’s results

In this post, we are going to see how we can work with the dataset and the data loader objects that we created in the previous post. Remember from the previous post, that we have two PyTorch objects, a Dataset and a DataLoader.

  • train_set
  • train_loader

We are now ready to see how to work with these objects, so let’s get to it.

PyTorch Dataset: Working with the training set

Let’s begin by looking at some operations we can perform to better understand our data.

Exploring the data

To see how many images are in our training set, we can check the length of the dataset using the Python len() function:

> len(train_set)
60000

This 60000 number makes sense based on what we learned in the post on the Fashion-MNIST dataset. Suppose we want to see the labels for each image. This can be done like so:

> train_set.train_labels
tensor([9, 0, 0, ..., 3, 0, 5])

The first image is a 9 and the next two are zeros. Remember from posts past, these values encode the actual class name or label. The 9 for example is an ankle boot while the 0 is a t-shirt.

shirt

If we want to see how many of each label exists in the dataset, we can use the PyTorch bincount() function like so:

> train_set.train_labels.bincount()
tensor([6000, 6000, 6000, 6000, 6000, 6000, 6000, 6000, 6000, 6000])

Class imbalance: Balanced and unbalanced datasets

This shows us that the Fashion-MNIST dataset is uniform with respect to the number of samples from each class. This means we have 6000 samples for each class. As a result, this dataset is said to be balanced. If the classes had a varying number of samples, we would call the set an unbalanced dataset.

Class imbalance is a common problem, but in our case, we have just seen that the Fashion-MNIST dataset is indeed balanced, so we need not worry about that for our project.

To read more about the ways to mitigate unbalanced datasets in deep learning, see this paper: A systematic study of the class imbalance problem in convolutional neural networks.

Accessing data in the training set

To access an individual element from the training set, we first pass the train_set object to Python’s iter() built-in function, which returns an object representing a stream of data.

With the stream of data, we can use Python built-in next() function to get the next data element in the stream of data. From this we are expecting to get a single sample, so we’ll name the result accordingly:

> sample = next(iter(train_set))
> print('len:', len(sample))
len: 2

After passing the sample to the len() function, we can see that the sample contains two items, and this is because the dataset contains image-label pairs. Each sample we retrieve from the training set contains the image data as a tensor and the corresponding label as a tensor.

Since the sample is a sequence type, we can use sequence unpacking to assigned the image and the label. We will now check the type of the image and the label and see they are both torch.Tensor objects:

> print('types:', type(image), type(label))
types: <class 'torch.Tensor'> <class 'torch.Tensor'>

We’ll check the shape to see that the image is a 1 x 28 x 28 tensor while the label is a scalar valued tensor:

> print('shapes:', image.shape, label.shape)
shapes: torch.Size([1, 28, 28]) torch.Size([])

We’ll also call the squeeze() function on the image to see how we can remove the dimension of size 1. This is review of course.

> print('squeezed:', image.squeeze().shape)
squeezed: torch.Size([28, 28])

Again, based on our previous discussion of the Fashion-MNIST dataset, we do expect to see the 28 x 28 shape for our image. The reason we see a 1 on the first dimension of the tensor is because the number of channels needs to be represented. Opposed to RGB images that have 3 color channels, grayscale images have a single color channel. This is why we have a 1 x 28 x 28 tensor. We have 1 color channel that has a size of 28 x 28.

Let’s plot the image now, and we’ll see why we squeezed the tensor in the first place. We first squeeze the tensor and then pass it to the imshow() function.

> plt.imshow(image.squeeze(), cmap="gray")
> print('label:', label)
label: tensor(9)
boot

We get back an ankle-boot and the label of 9. We know that the label 9 represents an ankle boot because it was specified in the paper that we looked at in the previous post .

Alright. Let’s see how to work with the data loader now.

PyTorch DataLoader: Working with batches of data

We’ll start by creating a new data loader with a smaller batch size of 10 so it’s easy to demonstrate what’s going on:

> display_loader = torch.utils.data.DataLoader(
    train_set, batch_size=10
)

We get a batch from the loader in the same way that we saw with the training set. We use the iter() and next() functions.

There is one thing to notice when working with the data loader. If shuffle=True, then the batch will be different each time a call to next occurs. With shuffle=True, the first samples in the training set will be returned on the first call to next. The shuffle functionality is turned off by default.

# note that each batch will be different when shuffle=True
> batch = next(iter(display_loader))
> print('len:', len(batch))
len: 2

Checking the length of the returned batch, we get 2 just like we did with the training set. Let’s unpack the batch and take a look at the two tensors and their shapes:

> images, labels = batch

> print('types:', type(images), type(labels))
> print('shapes:', images.shape, labels.shape)
types: <class 'torch.Tensor'> <class 'torch.Tensor'>
shapes: torch.Size([10, 1, 28, 28]) torch.Size([10])

Since batch_size=10, we know we are dealing with a batch of 10 images and 10 corresponding labels. This is why we went plural on the variable names.

The types are tensors like we’d expect. However, the shapes are different from what we saw with the single sample. Instead of having a single scalar value as our label, we have a rank-1 tensor with 10 values. The size of each dimension in the tensor that contains the image data is defined by each of the following values:

(batch size, number of color channels, image height, image width)

The batch size of 10, is why we now have a 10 in the leading dimension of the tensor, one index for each image. The following gives us the first ankle-boot we saw before:

> images[0].shape
torch.Size([1, 28, 28])

> labels[0]
tensor(9)

To plot a batch of images, we can use the torchvision.utils.make_grid() function to create a grid that can be plotted like so:

> grid = torchvision.utils.make_grid(images, nrow=10)

> plt.figure(figsize=(15,15))
> plt.imshow(np.transpose(grid, (1,2,0)))

> print('labels:', labels)
labels: tensor([9, 0, 0, 3, 0, 2, 7, 2, 5, 5])

Recall that we have the following table that shows the label mapping to class names below:

Index Label
0 T-shirt/top
1 Trouser
2 Pullover
3 Dress
4 Coat
5 Sandal
6 Shirt
7 Sneaker
8 Bag
9 Ankle boot

Building the model is next

We should now have a good understanding of how to explore and interact with Datasets and DataLoaders. Both of these will prove to be important as we begin building our convolutional neural network and our training loop. In fact, the data loader will be used directly inside our training loop.

neural network

Let’s keep moving as we are ready to build our model in the next post. I’ll see you there!

Description

Exploring and preparing data for neural network programming with PyTorch. We explore our training set, show images on a plot, and touch on oversampling. Check out the corresponding blog and other resources for this video at: http://deeplizard.com/learn/video/mUueSPmcOBc Code: https://www.patreon.com/posts/code-for-pytorch-21607032 Code files are available as a perk for the deeplizard hivemind. Check out the details regarding deeplizard perks and rewards at: http://deeplizard.com/hivemind ❤️🦎 Special thanks to the following polymaths of the deeplizard hivemind: Ruicong Xie Support collective intelligence, and join the deeplizard hivemind! Follow deeplizard: YouTube: https://www.youtube.com/deeplizard Twitter: https://twitter.com/deeplizard Facebook: https://www.facebook.com/Deeplizard-145413762948316 Steemit: https://steemit.com/@deeplizard Instagram: https://www.instagram.com/deeplizard/ Pinterest: https://www.pinterest.com/deeplizard/ Check out products deeplizard suggests on Amazon: https://www.amazon.com/shop/deeplizard Support deeplizard by browsing with Brave: https://brave.com/dee530 Support deeplizard with crypto: Bitcoin: 1AFgm3fLTiG5pNPgnfkKdsktgxLCMYpxCN Litecoin: LTZ2AUGpDmFm85y89PFFvVR5QmfX6Rfzg3 Ether: 0x9105cd0ecbc921ad19f6d5f9dd249735da8269ef References: Ted talk: https://youtu.be/clfFT1XOJKM Jeremy: https://youtu.be/O5F9vR2CNYI?t=1026 Recommended books on AI: The Most Human Human: What Artificial Intelligence Teaches Us About Being Alive: http://amzn.to/2GtjKqu Life 3.0: Being Human in the Age of Artificial Intelligence https://amzn.to/2H5Iau4 Playlists: Data Science - https://www.youtube.com/playlist?list=PLZbbT5o_s2xrth-Cqs_R9-us6IWk9x27z Machine Learning - https://www.youtube.com/playlist?list=PLZbbT5o_s2xq7LwI2y8_QtvuXZedL6tQU Keras - https://www.youtube.com/playlist?list=PLZbbT5o_s2xrwRnXk_yCPtnqqo4_u2YGL TensorFlow.js - https://www.youtube.com/playlist?list=PLZbbT5o_s2xr83l8w44N_g3pygvajLrJ- PyTorch - https://www.youtube.com/watch?v=v5cngxo4mIg&list=PLZbbT5o_s2xrfNyHZsM6ufI0iZENK9xgG Reinforcement Learning - https://www.youtube.com/playlist?list=PLZbbT5o_s2xoWNVdDudn51XM8lOuZ_Njv Music: Thinking Music by Kevin MacLeod Jarvic 8 by Kevin MacLeod YouTube: https://www.youtube.com/channel/UCSZXFhRIx6b0dFX3xS8L1yQ Website: http://incompetech.com/ Licensed under Creative Commons: By Attribution 3.0 License http://creativecommons.org/licenses/by/3.0/ #ai #pytorch #deeplearning