...It's back up again now. Looks like it was just a temporary disruption.

Using the same code:

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

raises a timeout error: TimeoutError: [WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

What would cause this?

Using the same code:

from tensorflow.examples.tutorials.mnist import input_data
mnist = input_data.read_data_sets("/tmp/data/", one_hot=True)

raises a timeout error: TimeoutError: [WinError 10060] A connection attempt failed because the connected party did not properly respond after a period of time, or established connection failed because connected host has failed to respond.

What would cause this?

The issue is that the website where MNIST is canonically hosted sometimes goes down. I think an ideal solution would be to find or set up a mirror, then fall back to it? Or just find a more stable source.

I also have the 10060 problem and it always happened. This seems like not the host problem. Do you know what caused this?

Reopening because others seem to have this problem.

I would suggest that if you encounter this problem, check http://downforeveryoneorjustme.com/http://yann.lecun.com/exdb/mnist/ just to make sure.

Right now it is:

(Sorry Yann!)

I cannot reach this website

Confirmed. I'm also having MNIST issues. Is there any alternative URL we can use for this which will still be compatible with existing MNIST tutorials. I'm relying on these as baselines for experiments, I'd rather keep the data and TF format the same.

Still down seems few days... Other related issue #8116

Same issue here, solved by downloading file from a different unofficial source. I think we should put in place an officially recognized mirror somewhere. Does anyone know if tensorflow checks a sort of signature for the dataset before using it? Or if a signature (e.g. md5) is available for the MNIST dataset so people can exchange it via unofficial circuits.

Also cannot access.

Closing as duplicate of #8126

@auserdude If you retrieve the .gz files from any other source (wayback machine?) and drop them in your data_dir then TF will just use them. It does not validate a checksum AFAIK. The bug I referenced has a suitable download URL.

ImportError: cannot import name 'model_fn' from 'tensorflow_estimator.python.estimator' (unknown location)

iget this this error when i tried mnsit alternative download..neither worked for me. i even downloaded mat file locally still get this error