mnist: CI has failed with 403 forbidden error

[tikkss]

CI (macos, ubuntu and windows) has failed:

• https://github.com/red-data-tools/red-datasets/actions/runs/9532177771/job/26273996707#step:8:1434

  Error: test: #each(MNISTTest::Normal::test): Net::HTTPClientException: 403: Forbidden: http://yann.lecun.com/exdb/mnist/t10k-images-idx3-ubyte.gz
  /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:163:in block (2 levels) in start_http' /Users/runner/hostedtoolcache/Ruby/3.3.3/arm64/lib/ruby/3.3.0/net/http.rb:2353:in block in transport_request'
  /Users/runner/hostedtoolcache/Ruby/3.3.3/arm64/lib/ruby/3.3.0/net/http/response.rb:320:in reading_body' /Users/runner/hostedtoolcache/Ruby/3.3.3/arm64/lib/ruby/3.3.0/net/http.rb:2352:in transport_request'
  /Users/runner/hostedtoolcache/Ruby/3.3.3/arm64/lib/ruby/3.3.0/net/http.rb:2306:in request' /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:149:in block in start_http'
  /Users/runner/hostedtoolcache/Ruby/3.3.3/arm64/lib/ruby/3.3.0/net/http.rb:1570:in start' /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:145:in start_http'
  /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:50:in block in download' /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:128:in block in synchronize'
  internal:kernel:187:in loop' /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:98:in synchronize'
  /Users/runner/work/red-datasets/red-datasets/lib/datasets/downloader.rb:34:in download' /Users/runner/work/red-datasets/red-datasets/lib/datasets/dataset.rb:38:in download'
  /Users/runner/work/red-datasets/red-datasets/lib/datasets/mnist.rb:49:in each' /Users/runner/work/red-datasets/red-datasets/test/test-mnist.rb:70:in each'
  /Users/runner/work/red-datasets/red-datasets/test/test-mnist.rb:70:in to_a' /Users/runner/work/red-datasets/red-datasets/test/test-mnist.rb:70:in block (3 levels) in class:MNISTTest'

• https://github.com/red-data-tools/red-datasets/actions/runs/9532177771/job/26273996776#step:8:1433
• https://github.com/red-data-tools/red-datasets/actions/runs/9532177771/job/26273996879#step:8:1451

Should we use a mirror site instead of original site?

See also:
oreilly-japan/deep-learning-from-scratch#81

[kou]

It seems that we should not always use https://ossci-datasets.s3.amazonaws.com/mnist/ . We can use it only when http://yann.lecun.com/exdb/mnist/ is not available:

pytorch/vision#4634 (comment)

AFAIK the author only authorized us to use the mirror in case the original server is unavailable for some reason.