canot load EMNIST dataset

[davidshen84]

/!\ PLEASE INCLUDE THE FULL STACKTRACE AND CODE SNIPPET

Short description
Failed to load the emnist dataset

Environment information

• Operating System: Linux

• Python version: 3.9

• tensorflow-datasets/tfds-nightly version: 4.9.4

• tensorflow/tf-nightly version: 12.6.1

• Does the issue still exists with the last tfds-nightly package (pip install --upgrade tfds-nightly) ?
  ✅

Reproduction instructions

import tensorflow_datasets as tfds

tfds.load("emnist", split=["train"])

If you share a colab, make sure to update the permissions to share it.

Link to logs

Expected behavior
The emnist dataset is loaded successfully.

Additional context

NonMatchingChecksumError: Artifact https://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip, downloaded to /root/tensorflow_datasets/downloads/itl.nist.gov_iaui_vip_cs_links_EMNIST_gzipi4VnNviDSrfd9Zju6qv40flc3wr22t8ldulNStS6tmk.zip.tmp.8cdbd18c3c7144529f0a2a11d1829c60/itl, has wrong checksum:
* Expected: UrlInfo(size=535.73 MiB, checksum='fb9bb67e33772a9cc0b895e4ecf36d2cf35be8b709693c3564cea2a019fcda8e', filename='gzip.zip')
* Got: UrlInfo(size=110.12 KiB, checksum='bfd529724d06f22872f32d6649561a57fd25ec17ea51d6f2ad24b96ea0519c34', filename='itl')
To debug, see: https://www.tensorflow.org/datasets/overview#fixing_nonmatchingchecksumerror

I tried to download the file directly using the link https://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip, but I got redirected to the NIST homepage. I think the link is outdated.

[marcenacp]

@davidshen84 Well spotted, thanks for opening the issue! It seems the URL (https://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip) now redirects to https://www.nist.gov/itl which causes the problem.

Did you find the actual link?

[davidshen84]

I cannot find any direct download link from the Internet. According to this page, https://www.nist.gov/itl/products-and-services/emnist-dataset, contacting the author is the only way to get the data set.

…

On Thu, 11 Apr 2024, 19:17 Pierre Marcenac, ***@***.***> wrote: @davidshen84 <https://github.com/davidshen84> Well spotted, thanks for opening the issue! It seems the URL ( https://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip) now redirects to https://www.nist.gov/itl which causes the problem. Did you find the actual link? — Reply to this email directly, view it on GitHub <#5356 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAQBTL6FMUGUOGH6XLJ6O3Y4ZIKDAVCNFSM6AAAAABF25B3B2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANBZGI3TAMBTHE> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[davidshen84]

Sorry, I just found out on that NIST page, the "original MNIST dataset" link points to the EMNIST dataset. 😅 Can you check if TF can still handle that file? Thank you

…

On Thu, 11 Apr 2024, 21:45 Xi Shen, ***@***.***> wrote: I cannot find any direct download link from the Internet. According to this page, https://www.nist.gov/itl/products-and-services/emnist-dataset, contacting the author is the only way to get the data set. On Thu, 11 Apr 2024, 19:17 Pierre Marcenac, ***@***.***> wrote: > @davidshen84 <https://github.com/davidshen84> Well spotted, thanks for > opening the issue! It seems the URL ( > https://www.itl.nist.gov/iaui/vip/cs_links/EMNIST/gzip.zip) now > redirects to https://www.nist.gov/itl which causes the problem. > > Did you find the actual link? > > — > Reply to this email directly, view it on GitHub > <#5356 (comment)>, > or unsubscribe > <https://github.com/notifications/unsubscribe-auth/AAAQBTL6FMUGUOGH6XLJ6O3Y4ZIKDAVCNFSM6AAAAABF25B3B2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANBZGI3TAMBTHE> > . > You are receiving this because you were mentioned.Message ID: > ***@***.***> >

[davidshen84]

This should be the new emnist dataset URL: https://biometrics.nist.gov/cs_links/EMNIST/gzip.zip

[minchan0410]

Did you fix the error? If you have solved it, can you tell me how? I also get the same error :(

[davidshen84]

No. They just need to update the URL. I think you can manually download the archive and put it in the download folder. TF will skip the downloading, thus skip this bug.

…

On Wed, 8 May 2024, 00:36 minchan0410, ***@***.***> wrote: Did you fix the error? If you have solved it, can you tell me how? I also get the same error :( — Reply to this email directly, view it on GitHub <#5356 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAAQBTO5CTBFCW3G4MBNVILZBDRFPAVCNFSM6AAAAABF25B3B2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDAOJYGU2TQOBYGA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[ccl-core]

Hello, #5401 which should have solved the issue is now merged!
Starting from tomorrow, the change will be available in tfds-nightly.

[minchan0410]

Thank you both for letting us know. It was helpful!! :)