HTTP Error when trying to download MNIST data

user_3616293 0
Viewed 9237

I am using Google Colab for training a LeNet-300-100 fully-connected neural network on MNIST using Python3 and PyTorch 1.8.

To apply the transformations and download the MNIST dataset, the following code is being used:

# MNIST dataset statistics:
# mean = tensor([0.1307]) & std dev = tensor([0.3081])
mean = np.array([0.1307])
std_dev = np.array([0.3081])

transforms_apply = transforms.Compose([
    transforms.ToTensor(),
    transforms.Normalize(mean = mean, std = std_dev)
    ])

which gives the error:

Downloading http://yann.lecun.com/exdb/mnist/train-images-idx3-ubyte.gz to ./data/MNIST/raw/train-images-idx3-ubyte.gz --------------------------------------------------------------------------- HTTPError Traceback (most recent call last) in () 2 train_dataset = torchvision.datasets.MNIST( 3 root = './data', train = True, ----> 4 transform = transforms_apply, download = True 5 ) 6

11 frames /usr/lib/python3.7/urllib/request.py in http_error_default(self, req, fp, code, msg, hdrs) 647 class HTTPDefaultErrorHandler(BaseHandler): 648 def http_error_default(self, req, fp, code, msg, hdrs): --> 649 raise HTTPError(req.full_url, code, msg, hdrs, fp) 650 651 class HTTPRedirectHandler(BaseHandler):

HTTPError: HTTP Error 503: Service Unavailable

What's wrong?

pytorch python
7 Answers

I was having the same 503 error and this worked for me

!wget www.di.ens.fr/~lelarge/MNIST.tar.gz
!tar -zxvf MNIST.tar.gz

from torchvision.datasets import MNIST
from torchvision import transforms

train_set = MNIST('./', download=True,
transform=transforms.Compose([
transforms.ToTensor(),
]), train=True)


test_set = MNIST('./', download=True,
transform=transforms.Compose([
transforms.ToTensor(),
]), train=False)
user_126047460

There has been a lot of trouble with the MNIST hosted on http://yann.lecun.com/exdb/mnist/ therefore pytorch got permission and hosting it now on amazon aws.

Unfortunately, the fix is only available in the nightly build (Here you can find the fixed code. )

A hot fix I found useful is:

from torchvision import datasets
new_mirror = 'https://ossci-datasets.s3.amazonaws.com/mnist'
datasets.MNIST.resources = [
   ('/'.join([new_mirror, url.split('/')[-1]]), md5)
   for url, md5 in datasets.MNIST.resources
]
train_dataset = datasets.MNIST(
   "../data", train=True, download=True, transform=transform
)

Update: According to torch vision issue 3549 this will be fixed in the next minor release

user_34115170

This problem has been solved in torchvision==0.9.1 according to this. As a temporary solution, please use the following workaround:

from torchvision import datasets, transforms
datasets.MNIST.resources = [
    ('https://ossci-datasets.s3.amazonaws.com/mnist/train-images-idx3-ubyte.gz', 'f68b3c2dcbeaaa9fbdd348bbdeb94873'),
    ('https://ossci-datasets.s3.amazonaws.com/mnist/train-labels-idx1-ubyte.gz', 'd53e105ee54ea40749a09fcbcd1e9432'),
    ('https://ossci-datasets.s3.amazonaws.com/mnist/t10k-images-idx3-ubyte.gz', '9fb629c4189551a2d022fa330f9573f3'),
    ('https://ossci-datasets.s3.amazonaws.com/mnist/t10k-labels-idx1-ubyte.gz', 'ec29112dd5afa0611ce80d1b7f02629c')
]

# AND the rest of your code as usual for train and test (EXAMPLE):
batch_sz = 100
tr_ = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])
# MNIST
train_dataset = datasets.MNIST(
    root='./dataset', 
    train=True, 
    transform=tr_,  
    download=True
)

test_dataset = datasets.MNIST(
    root='./dataset', 
    train=False, 
    transform=tr_  
)
# DataLoader
train_loader = torch.utils.data.DataLoader(
    dataset=train_dataset,
    batch_size=batch_sz,
    shuffle=True 
)

test_loader = torch.utils.data.DataLoader(
    dataset=test_dataset,
    batch_size=batch_sz,
    shuffle=False 
)
user_54370900

you can try this:

from sklearn.datasets import fetch_openml
mnist = fetch_openml('mnist_784', data_home=".")

x = mnist.data
x = x.reshape((-1, 28, 28))
x = x.astype('float32')

y = mnist.target
y = y.astype('float32')
user_100496460

for PyTorch 0.4.0 in udacity notebooks.

The solution is inspired by the above solution.

new_mirror = 'https://ossci-datasets.s3.amazonaws.com/mnist'
datasets.MNIST.urls = [
   str('/'.join([new_mirror, url.split('/')[-1]]))
   for url in datasets.MNIST.urls
]
transform = transforms.Compose([transforms.ToTensor(),
                              transforms.Normalize((0.5,), (0.5,)),
                              ])
user_101403790 
import tensorflow as tf
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()

use that

user_122972370

You did nothing wrong. It is problem of the platform where data is hosted. Using Pytorch you can download MNIST using below code

import torch
import torchvision
from torchvision.datasets import MNIST

# Download training dataset
dataset = MNIST(root='data/', download=True)

The above MNIST wrapper in Pytorch datasets would try many possible places where data is available. After running the code you can see that first it tries to download from Yan Le Cun site but fails to download from there and fall back to other possible options.

Potential Cause: The Yan LeCun Site is missing an updated SSL certificate, so some methods of downloading files do consider about this security measure and some doesn't.

user_155623880
Related