How to upload our dataset using pytorch when it is not present in torchvision?

Question

I am trying to upload my dataset(SWELL-KW) instead of MNIST in "Hybrid quantum-classical Neural Networks with PyTorch and Qiskit" provided by IBM qiskit but it says "AttributeError: module 'torchvision.datasets' has no attribute 'SWELL'. I could not find the dataset in torch vision. Is that the reason? If I try to give the path of dataset which is on my system, it indicates unicode error. Thank You.

Answer 1

Yes, since the dataset does not exist in torchvision, trying to load it through torchvision will produce this error. Instead, save your SWELL-KW dataset as a .csv file, read it in, and convert the train/test data vectors, and labels each as their own PyTorch tensor.

For example, let's say I have my own custom dataset that consists of coordinate pairs and associated labels. I save the data into a file myData.csv such that the first column contains the x-coordinates, the second column contains the y-coordinates and the third column contains the labels. First, we load data row by row,

import csv
import torch

Xs = []
Ys = []

with open('myData.csv', mode ='r') as data:
    csvFile = csv.reader(data)
    for row in csvFile:
        Xs.append([float(row[0]), float(row[1])])
        Ys.append(float(row[2]))

and then we convert each of our lists to PyTorch tensors,

X_tensor = torch.FloatTensor(Xs)
Y_tensor = torch.FloatTensor(Ys)

Your data is now ready to use. For more complex datasets or specific use cases you may need to create a custom dataset class.